This book examines an increasingly important phenomenon for competitiveness and innovation in industry: namely, the growing use of scientific principles in industrial research. Industrial innovation still arises from systematic trial-and-error experiments with many designs and objects, but these experiments are now being guided by a more rational understanding of phenomena. This has important implications for market structure, firm strategies, and competition. Science and innovation focuses on the pharmaceutical industry. It discusses the changes that the notable advances in the life sciences in the 1980s have brought to the strategies of drug companies, the organization of their internal research, their relationships with scientific institutions, the division of labor between large pharmaceutical firms and small research-intensive suppliers, the productivity of drug discovery, and the productivity of R&D.
Science and innovation
Science and innovation The US pharmaceutical industry during the 1980s Alfonso Gambardella University of Urbino
CAMBRIDGE
UNIVERSITY PRESS
Published by the Press Syndicate of the University of Cambridge The Pitt Building, Trumpington Street, Cambridge CB2 1RP 40 West 20th Street, New York, NY 10011-4211, USA 10 Stamford Road, Oakleigh, Victoria 3166, Australia © Cambridge University Press 1995 First published 1995 A catalogue recordfor this book is available from the British Library Library of Congress cataloguing in publication data applied for ISBN0 521 45118 3 hardback Transferred to digital printing 2004
SE
To Anjana
Contents
List of List of tables Preface List of abbreviations
figures
page x xi xii xvi
1
Introduction
2
Science and innovation in pharmaceutical research
18
3
Economic implications of greater scientific intensity in drug research
42
In-house scientific research and innovation: case studies of large US pharmaceutical companies
82
Scientific research and drug discovery: an econometric investigation
106
A model of the innovation cycle in the pharmaceutical industry
124
4 5 6 7
8
1
Complementarity and external linkages: the strategies of large firms in biotechnology
146
Conclusions
161
Notes References Index
168 179 195 IX
Figures
2.1 2.2 4.1 4.2 6.1 6.2
Stages in the development of new drugs Number of new drugs commercialized worldwide, 1987-91 Smithkline's R&D/sales ratio, 1959-88 Scientific papers published by company scientists: Smithkline, Eli Lilly, and Merck, 1964-88 Estimated lag structure of R&D Growth rate of market value
page 19 40 96 97 138 141
Tables
2.1 2.2 3.1
3.2 3.3 3.4 3.5 4.1 4.2 4.3 4.4 5.1 5.2 5.3 5.4 5.5 6.1 6.2 6.3 6.4 6.5 6.6 7.1 7.2
US drug development costs by stage page 20 Top ten pharmaceutical products in 1991 24 Agreements and other relationships (in biopharmaceuticals) between large US pharmaceutical firms and universities or other research institutions in the 1980s 49 Ciba-Geigy's network of agreements 63 Hoffmann La Roche's network of agreements 66 Merck's network of agreements 70 Pfizer's network of agreements 73 1986 pharmaceutical sales and 1983-88 average R&D/ sales ratio of the eight firms in the case studies 84 Merck: major new drugs in the market or in the pipeline, late 1980s 85 Eli Lilly's drugs with 1989 sales of at least $100 million 89 Eli Lilly's research alliances in the early 1990s 91 Descriptive statistics: levels 113 Descriptive statistics: logs 113 Estimated parameters, equations (7)-(8) 115 Estimated parameters, equations (7)-(8) with structural differences in parameters, 1968-79 and 1980-91 116 Hypothesis testing 118 Descriptive statistics: levels 135 Descriptive statistics: growth rates - log (yit/yit-1) 135 Estimated parameters, equations (3), (5), (6), and (7) 136 Estimated lag structure of R&D 137 Estimates of the shock parameters and shock variances 139 Growth rate of market value 140 Definition of the variables used in the empirical analysis 153 Descriptive statistics 153
xii
List of tables
7.3 7.4
Results from negative binomial estimation of AWF, AWU, PRL, and PRM Correlations among the standardized residuals from negative binomial estimation of AWF, AWU, PRL, and PRM in table 7.3
156 157
Preface
This book deals with an important feature of technical change in industry towards the end of the twentieth century. The relationships between science and innovation are as old as the beginnings of capitalism, if not older. Yet, the 1980s and early 1990s have witnessed a considerable increase in the use of abstract, scientific conceptualization in the "for-profit" research sector. Moreover, impressive developments in computational power and softwarebased simulation have improved significantly the technology of experimentation. Altogether, these advances are bringing about prominent changes in the nature of industrial research and the innovation process. Although in different ways and to different degrees, this trend is affecting all high-tech industries. But in order to provide extended evidence of this pattern, one would have to deal with the many idiosyncratic features that characterize and distinguish the relationships between science and innovation in different sectors. This is why, in this book, I decided to focus on the pharmaceutical industry, wherein, for several reasons, the 1980s attested a clear shift from largely empirical industrial research processes (based on trial and error of many compounds) to a more rational search for innovation, based on effective use of scientific knowledge and computerized research technologies. I shall emphasize several times that the pharmaceutical industry is special. Nonetheless, it could well exhibit, in ways that are more apparent than elsewhere, evolutionary patterns that are important in other high-tech (and non high-tech) industries as well. If so, we could be on the verge of notable transformations in the "technology of technological change," with related implications for competition, division of labor, relationships with non-profit scientific institutions, and the like. While I do not address these broader questions here, I leave them for speculation by the reader, and to future economic research. There are many people I would like to acknowledge. I am especially indebted to Nathan Rosenberg, who made me discover the subtleties of the xiii
xiv
Preface
economics of innovation, encouraged me to pursue the topic of this work, and provided many insightful comments. Along with Nate, I want to mention Rina's dear and warm friendship. Tim Bresnahan and Paul David have been sources of many valuable suggestions. Paul is an economic historian with, uncommonly, a technical background, whilst Tim is a technical economist with a marvelous sense of economic history. Among other things, their versatility was of tremendous help infindinga technical basis for historical facts, and in giving historical meaning to technical results. I have now developed such close personal and professional linkages with Ashish Arora that it was extremely difficult to disentangle things that I had not discussed with him or worked out in previous papers with him, or that we shall work out together in the future. Although I refer to our work in the text, his influence on my thinking about these issues extends to many other parts of the book. Moreover, chapter 7 is based upon joint work with him. Ed Steinmueller provided penetrating comments both in writing and in discussion. His deep understanding of technologies, and his attention to economic implications, have constituted a model that I have tried to imitate. Gianni Cozzi and Sergio Vacca, along with Barbara Di Bernardo and Enzo Rullani, have greatly influenced my way of approaching the study of firms, technologies, and industries. I learned many things about the pharmaceutical industry and the broadly defined chemical sector from Ralph Landau. Bronwyn Hall provided useful comments on an earlier draft of chapter 6, and kindly supplied me with the data from the NBER-Compustat File. I also discussed topics related to this work with Shane Greenstein, Roger Noll, John Pencavel, Geoffrey Rothwell, Roland Sturm, and Frank Wolak. Among my Italian friends and colleagues, I have benefited from discussion with Salvo Torrisi and Antonello Zanfei, and with Franco Malerba, Luigi Orsenigo, and Alina Rizzoni. Paolo Barbanti, Nicoletta Buratti and Raffaella Delia Valle helped me in improving my understanding of the pharmaceutical industry. Francesco Delia Valle, a top manager and now an entrepreneur in this industry, deserves a special mention. He discussed and shared many views expressed in this book, and made me aware that they were in accord with the "pragmatic" expertise of some drug industry executives. Italy needs many more entrepreneurs like him. I wish to thank Steven Gass of the Terman Engineering Library at Stanford University, and Wanda Cantarello of Lifegroup SpA, for conducting the on-line search of the 1980-91 pharmaceutical patent data, and Francis Narin and Steven Kimberly of CHI/Computer Horizons Inc. for providing me with data on the scientific publications of pharmaceutical
Preface
xv
companies. Amanda Feuz patiently went through my manuscript to smooth out my crunchy English. My wonderful wife and companion in life, Monica, and my family have been sources of endless encouragement. The book is dedicated to my daughter.
I gratefully acknowledge financial support for this research from: the Technology and Economic Growth Program, Stanford University, USA (1988-89); the Lynde and Harry Bradley Foundation, USA (1990); the Italian Ministry of Scientific Research (MURST), which funds 60 percent of the University of Urbino, Italy (1991-93); the Centro Nazionale delle Ricerche (CNR), Progetto Finalizzato "Processi di Internazionalizzazione delle Impresse," contract no. 92.02634.PF73-115.10801, Italy (1992); the EC-DG XII Program "Human Capital and Mobility," contract no. ERISCHRXCT920002 (1993-94).
Abbreviations
AHP FDA IND NBER NDA OTC
XVI
American Home Products Food and Drug Administration Investigational New Drug Exemption National Bureau of Economic Research New Drug Application over the counter
Introduction
Science and innovation in industrial research In the Wealth of Nations, Adam Smith stated that "improvements in machinery" are sometimes made by "philosophers and men of speculation," "who . . . are often capable of combining together the powers of the most distant and dissimilar objects" (Smith, 1982 [1776], p. 114). Marx, in chapter 15 of Das Kapital, and more recently Kuznets (1966) emphasized that a major peculiarity of capitalism is the application of science to problems of industry. This is consistent with general historical evidence. Thermodynamics supplied the scientific basis of the First Industrial Revolution (Landes, 1969). The works of Faraday and Maxwell on electricity and magnetism, and Hertz's discovery of radio waves, laid the ground for the electricity-based industries and the Second Industrial Revolution (Landes, 1969 and 1991). In this century, Wallace Carothers and William Shockley clarified important theoretical aspects about polymer chemistry and solid-state physics which spurred the growth of the synthetic fiber and electronics industries (Nelson, 1962; Braun and MacDonald, 1978; Smith and Hounshell, 1985; Hounshell and Smith, 1988). But, in spite of these (and other) examples, science, as a body of "generic" and systematized knowledge, has rarely offered more than a general and indirect framework for innovation. Most innovations and productivity improvements in industry have stemmed from highly empirical procedures based on trial-and-error experiments. Moreover, innovations have relied on fairly "old" science, and the application of fundamental knowledge to problems of industry has occurred with relatively long time-lags (Rosenberg, 1985; Mowery and Rosenberg, 1989). The empirical character of industrial research reached its climax with the institutionalization of R&D in large corporations during the postwar years of this century. As Schumpeter taught us, big firms could afford the high 1
2
Introduction
set-up costs of large in-house research laboratories. They were able to test several modifications of product and process design, and carry out a great many experiments in routine and efficient ways. The hit-and-miss nature of business research arose from the fact that companies could make only limited use of knowledge to scale up and transfer experimental results. There was a great deal of uncertainty about whether experimental results obtained for "small" prototypes or in the laboratory would hold for larger objects or at larger scales. Similarly, it was difficult to relate the results of a given test to other experiments and phenomena (David, Mowery, and Steinmueller, 1992; Arora and Gambardella, 1993a and 1994b). Hence, industrial scientists and engineers had to carry out many careful and systematic experiments each time they wanted to evaluate the properties and performance of a certain device, and each time they wanted to scale up an object or process from smaller-scale tests. In aeronautics, for example, engineers for many years have used wind tunnels to simulate the flight of airplanes. Wind tunnels have enabled them to observe the performance of various modifications of a specific design. Because of the lack of a general theory of aerodynamics, long and costly experiments have been the only reliable way to develop and improve aircraft (Mowery and Rosenberg, 1982; Rosenberg, 1982; Vincenti, 1990). Similarly, chemical engineers have lacked a comprehensive theory of the relationships between catalysts and chemical reactions. In the chemical and petrochemical industries, the development of new catalysts has rested on laborious experimentation with a large number of molecules to find the few that could satisfactorily raise the efficiency of a certain reaction (Cusumano, 1992). In manufacturing, production engineers have lacked a thorough understanding of how production processes work. Most process improvements have stemmed from intentional, methodical variations of process parameters (pressure, temperature, etc.) to seek more efficient combinations of the parameters themselves, or to determine how a process that works at a small scale could function at larger scales as well (Mansfield etal., 1977; Steinmueller, 1987). An important thrust of this book is to argue that this state of affairs is changing. This is largely related to spectacular developments in many scientific disciplines, along with impressive progress in computational capabilities and instrumentation. While many of these advances originated around the Second World War, or even earlier, they accelerated considerably during the 1970s and, especially, the 1980s. In addition, the lags between scientific discoveries and industrial applications appear to have shortened.1 Important advances in theoretical chemistry, life sciences, particle and solid-state physics, to name a few, are enhancing our understanding of phenomena that occur at the level of "nano"-structures, or in inappreciable
Science and innovation in industrial research
3
fractions of time (e.g., chemical reactions that take less than one-millionth of a second). Scientists and engineers can then study the "causes" of many events that could not be comprehended without understanding the behavior of the infinitesimal components of objects and systems. They can examine the aggregation of micro-structures to form materials with desirable properties, or they can study optical and electrical conductivity to generate major advances in electronics, information systems, and in many other technologies. Relatedly, advances in computational capabilities, computer science, and corresponding theoretical tools (e.g., chaos theory, genetic algorithms, neuronal networks, and artificial intelligence) are making it possible to understand very complex problems, like the flight of airplanes, the dynamics of mechanical and electro-mechanical systems, or even biological systems. These developments in theory owe a great deal to complementary advances in the technology of research. New and more powerful instruments have enabled scientists and engineers to "observe" infinitesimal objects. Using X-ray crystallography, nuclear magnetic resonance, the scanning electron microscope, laser, and other new magnetic- or opticalbased techniques, they can "look" at the arrangement of atoms and molecules in exceptionally small clusters (De Solla Price, 1984; Lederman, 1984; Rosenberg, 1992). This is important, as the geometry of atoms and molecules governs the properties of matter (CSE&PP, 1983; Baker, 1986; PSI&TA, 1986). Also, using advanced instrumentation techniques - like ion implantation or, more recently, synchroton radiation - they can intervene at atomic and sub-atomic levels to obtain selected alterations of matter, and hence control specific properties of materials like strength or conductivity (PSI&TA, 1986; Steinmueller, 1987; David, Mowery, and Steinmueller, 1992). The possibility of observing and intervening on microscopic objects has significantly helped the development of theories. Not only can theories be tested, but, more importantly, they can be perfected and corrected from observation. This is not a novel feature of either industrial or scientific research. However, because one can now observe things that could not previously be observed, it is possible to improve theories that could not be supported by visual inspection. Empirical observation also reveals new "facts," issues and problems that, in turn, stimulate the progress of theories. In addition, advances in computer power (both hardware and software) are making it possible to perform efficient simulations of experiments, with implied time and cost advantages over physical experiments. One can model the behavior of infinitesimal objects or highly complex systems, and observe them on computers. Computer simulation is now widely used in
4
Introduction
aeronautics, automobiles, chemicals, and - as we shall see in this book Pharmaceuticals. For instance, it may take only a few minutes to test the resistance of a new material in car crashes. The computer simulates the crash, and measures the deformation of the body at different strengths of collision. It may even elaborate the data, and suggest directions for improvements (e.g., which parts of the body need to be reinforced). Moreover, like instrumentation, computer models hint at issues and problems that suggest directions for new developments of basic knowledge. In short, theoretical advances, combined with powerful instrumentation and computational capabilities, are encouraging the diffusion of a new approach to industrial research. Instead of "blind" (physical) experiments to find what may work, industrial scientists and engineers increasingly test hypotheses using sophisticated instruments, and simulate processes on computers. This is not to imply that industry can now do without physical experiments. The latter are still necessary because the complexity of many phenomena cannot be fully comprehended using theory or simulation. Also, the new trend does not imply that innovation now follows a linear, "cascade" model, from basic research to development and commercialization. It still depends on complex feedbacks among different stages of the innovation cycle (Kline and Rosenberg, 1986). However, relevant information for innovation, from whichever sources it arises (research, production, markets), can be cast in more general frameworks. Furthermore, the blend of new powerful theories, instruments and simulation, helps in designing more efficient physical tests. Scientists and engineers can exclude from the physical experiments a number of objects and events that are inconsistent with a priori hypotheses (Arora and Gambardella, 1994b). The literature on science as a source of economic value Private appropriability of scientific research The increasing use of fundamental knowledge and of the tools of experimental science in industrial research poses important economic questions. Is the trend just described a widespread one in industry, or is it confined to very few firms at the frontier of science and technology? What implications are there for the organization of company R&D? And for the size of the innovating firms? How is it going to affect the relationships between companies and external parties (research institutions, other companies, etc.)? What is the role of public knowledge and scientific institutions? What implications are there for competition and market structure? Before addressing these issues in the context of the pharmaceutical industry, it may be useful to review the economic literature on the
Science as a source of economic value
5
relationships between science and innovation. This will help in focusing our discussion, and in structuring our questions more carefully. The seminal contributions of Nelson (1959) and Arrow (1962) are a natural starting point. Their well-known conclusion is that profit-seeking agents invest less than optimally in scientific research. Basic research produces information. But information is a peculiar commodity in that it can be re-produced at zero cost. Any buyer of the information, or anyone who happens to have it, can become a producer of it. This destroys the monopoly of the rent-seeking agent who invested in producing the information. Patents and intellectual property rights restore, in part, the incentives to perform fundamental research. However, the low appropriability of basic science and technology explains why the bulk of this activity is carried out by institutions not motivated by economic profits (universities, government research centers, etc.), and why it has to be sustained by public funds (Mowery, 1983). The low appropriability of science has been criticized on various grounds. Many authors have argued that scientific research, and particularly the ability to utilize knowledge for industrial purposes, depend on tacit skills, learning, and organizational routines.2 These are firm-specific factors that cannot be easily transferred between companies. Firms with better research skills and assets, and possibly with better downstream resources for commercialization, can then secure more effectively the benefits of their investment in basic science and technology, and they have greater incentives to perform fundamental research. The growing application of scientific and technological principles to industry further weakens the low appropriability argument. As basic knowledge exhibits more direct effects on innovation, and shorter application lags, companies can embody more thoroughly and more rapidly their research outcomes in new products and processes. To the extent that they can appropriate innovations through patents, first-mover advantages or downstream assets, they can also shelter more productively their investments in basic science and technology. But the incentives of private firms to invest in fundamental research do not depend only on the possibility of appropriating and utilizing one's own research outcomes. Another important reason is that in-house scientific and technological capabilities are critical for monitoring and utilizing external knowledge. Rosenberg (1990) argued that, although a public good, science is by no means a free good. (See also Pavitt, 1991.) The ability of firms to exploit public science will be greater if they perform similar research themselves. Investments in research are thus necessary to comprehend and evaluate external information. (See also Arora and Gambardella, 1994a.) Cohen and Levinthal (1989) argued that R&D has two "faces." Apart
6
Introduction
from generating innovations, it helps in absorbing knowledge generated by others. If the second face of R&D is important, variables that affect the ease with which firms learn from the pool of outside knowledge ought to influence their technological opportunities and appropriability conditions. In turn, this would influence their incentives to invest in R&D. Cohen and Levinthal developed a model of the accumulation of knowledge capital. They assumed that the knowledge capital of firms depends on the pool of outside knowledge, and on technological opportunity and appropriability factors. They used data from the Federal Trade Commission's Line of Business Program on business-unit R&D expenditures. They found that the influence of technological opportunity factors and appropriability conditions on R&D spending is affected by variables that influence the ease of learning. This suggests that the second face of R&D is important in encouraging firms to invest in research. Although they used data on total R&D expenditures, Cohen and Levinthal speculated about the incentives of firms to perform basic research.3 Their analysis leads to predictions about the incentives for firms to conduct scientific research. Factors affecting the firms' opportunities of learning ought to influence investments in in-house basic research. In this respect, the acceleration of scientific developments, and the more rapid application of science to industry, ought to encourage private outlays in science as a means of absorbing the rising external opportunities. The educational and research functions of universities As industrial research depends more substantially on formalized knowledge, the educational function of universities will become more important. Gibbons and Johnston (1974) used a sample of 30 innovations and 887 "units" of information relevant to generating the innovations. They classified the units of information according to their source (trade literature, technical literature, scientific literature, personal contacts with scientists, customers, suppliers, etc.) and content (theories, design-based information, information about materials, etc.). They found that people with university education adopt a more general approach, solve more general problems, and rely on a wider pool of information drawn largely from sources external to their company. People without university education solve more specific problems, and rely primarily on knowledge and experience gained within their company. The need to tackle problems from a more universal perspective, and to take advantage of external information implies, therefore, that companies will hire more people with a solid academic background. Using data from a questionnaire survey with 650 R&D managers in
Science as a source of economic value
7
various industries, Nelson (1986, 1988, and 1994) studied whether the business community places greater value on the research or the educational function of universities (see also Levin et ai, 1987). The R&D managers were asked to rank the relevance of various fields of basic and applied science to technical change in their line of business. They were also asked to rank the relevance of university research in those fields. Almost every field of science received a high score. In contrast, the scores on university research in each field were significantly smaller. Nelson interpreted this result as evidence that, whilst university training is important, industry is less keen on academic research. Computers, drugs, and new materials were notable exceptions in that R&D managers in these industries gave high scores to university research as well. A recent report which summarized interviews with seventeen senior US company officials in a number of industries reinforces Nelson's conjecture (GUIRR, 1991). The officials emphasized the importance of university training. They also stated that universities provide in-depth understanding of new scientific and technological ideas. However, they remarked that, apart from a few cases, such as biotechnology, academic research provides little contribution to the vast and important downstream development activities that lead to commercially useful innovations. This is not contradictory to the argument that science is playing a more important role for innovation. The point is not whether science, and particularly academic research, generates straight commercial applications, but whether it produces new technological opportunities. Nelson (1986) regressed industry R&D intensity against the contribution of university research to innovation and other variables. He measured the contribution of university research by the score given by the R&D managers in his sample to the importance of university research. He found that this measure is positively correlated with R&D intensity. However, when regressing measures of technical progress against R&D intensity and the university research score (and other variables), R&D showed a positive and significant effect, whilst university research did not. This suggests that university research enhances technological opportunities and the productivity of private R&D, which in turn induces greater R&D spending in industry. Jaffe (1989) and Mansfield (1991) also examined the importance of university research for innovation and industry R&D. Using US data, Jaffe found that university research has a positive and significant effect on corporate patents and industry R&D. The latter result reinforces Nelson's conjecture that university research influences innovation through industry R&D. Jaffe also found that geographical proximity increases the strength of the effect of university research on corporate patents. The contribution
8
Introduction
of university research is greater if industry and university scientists can interact more easily.4 Using a sample of innovations from seventy-six large US firms in seven industries (information processing, electrical, chemicals, instruments, drugs, metals and oil), Mansfield found that, in the period 1975-85, about 19 percent of the new products and 15 percent of the new processes could not have been developed without substantial aid from recent academic research. He also found that the average time-lag between academic discoveries and industrial applications was seven years. Finally, although based on conservative assumptions, Mansfield estimated a high social rate of return of academic research (28 percent). Both Mansfield and Jaffe found that the effects of academic research are greater in the drug industry than in the mechanical and electrical industries. The scientific and technological communities Increasing scientific intensity of technical progress calls for a more rigorous economic investigation of the nature and behavior of academic institutions. In his early study of the transistor, Nelson (1962) had already pointed out that "[t]he incentives of the scientists are much more complex than the incentives of the entrepreneur of economic theory . . . The furthering of a reputation . . . and the satisfaction of intellectual curiosity are certainly as important . . . dimensions of the research worker's goal as are financial advance and status within an organization" (Nelson, 1962, p. 573; see also Nelson, 1982). In later works, Nelson (1988 and 1990) further discussed the importance of universities and research institutions in the diffusion of knowledge, and their contribution to the rising "socialization" of knowledge and research in modern capitalism. Nelson and Rosenberg (1993) studied the origins and development of American universities. They showed that, initially, American universities performed very applied research tasks. They were suppliers of research services to local industry (e.g., chemical tests and the development of special techniques). On some occasions, they developed entirely new engineering disciplines to serve special industry needs (e.g., chemical engineering at MIT between the two wars - see also Landau, 1990). Only after World War II, with the growth of Federal funding to universities, did American academia move significantly into basic research. Nelson and Rosenberg's work is important in that it reminds us that the contribution of many universities to technical progress does not take the form of fundamental discoveries, but of more applied research tasks and experiments. Following previous works in the sociology of science, Dasgupta and David (1987 and 1994) offered a comprehensive economic investigation of
Science as a source of economic value
9
the different social behavior and institutional construct of the scientific community compared with those of rent-seeking "technologists." 5 Well before the beginnings of capitalism, scientists needed "patrons" who could support their activities. But in order to gain visibility so as to be employed in the courts of kings and princes, they had to publicize their abilities (see also David, 1991). This prompted the social organization of science that we observe today, which encourages disclosure of findings. Today's scientists diffuse their results to raise their public recognition and peer-group esteem. Apart from personal satisfaction, reputation increases their ability to raise funds for future research (both from government and industry), which in turn enhances their research skills (and hence human capital), as well as their salaries and prestige. Public reputation also enables them to obtain personal consultancies from companies, with implied pecuniary rewards. In contrast, the technological community seeks to appropriate the rents that can be derived from research. Disclosure would undermine their opportunities to gain such rents. Firms then encourage secrecy and appropriation of findings through patents and other means. Dasgupta and David (1994) also investigated several economic aspects of scientific institutions, such as the mechanisms for allocation of funds, co-operation and competition among scientists, the interactions between scientific institutions and companies, and the effects of industry funding on academic research. The distinction between scientists and technologists raises some questions. For one, can we really associate the technological community with firms and the scientific community with universities and other nonprofit research institutions? Companies like IBM, General Electric, and Du Pont maintain some in-house research facilities that are organized in ways that are similar to academic departments. They employ scientists who publish, participate in conferences, and behave almost like their colleagues in universities. (Some of them even win Nobel prizes!) Many small innovative firms in high-tech sectors (e.g., electronics, software, and biotechnology) thrive largely because of their informal research organizations and academic linkages. At the same time, as suggested earlier, many universities perform very applied research. They normally produce research services and consultancy to local industry rather than purely academic publications. They sometimes restrain diffusion to preserve confidentiality for their clients. A good fraction of their activities feed the domain of research controlled by agents who seek appropriation of rents and secrecy rather than public knowledge. Dasgupta and David's distinction, however, provides a useful framework for our discussion. The issue is precisely whether the growing applicability of science to industry, and the corresponding greater inter-
10
Introduction
dependence of firms and the scientific community, could affect the ways in which the two institutions have conducted their research for many years. Will the need for greater scientific creativity and more intense linkages with external science lead to greater openness of company research? As public research budgets shrink, and universities seek more funds from industry, will industry push for greater secrecy in academic research, with implied restrictions on the socially desirable circulation of knowledge? Will companies force academia to shift from "generic" research towards more narrow, business-oriented goals? The possibility of greater restrictions on academic research is especially intriguing. Dasgupta and David discussed this issue in a historical perspective. Open science predated capitalism. It was made possible by the social organization of science based on the patronage of kings and princes. This furthered the evolution of modern industry by ensuring widespread diffusion of knowledge. In our own day, governments are the patrons of science. They have chosen to support science through public funds in recognition of the fact that the long-run social benefits of open science could conflict with the short-run goals of private companies. But if public support to science diminishes, firms may become the new patrons of science. Because of their short-run, private objectives, they may impose restrictions on diffusion, thereby undermining the very same factor that has encouraged their longrun growth. The distinction between scientists and technologists hints at another important difference between research in academic institutions and firms. (See also Arora and Gambardella, 1993b and 1994b.) Scientific prestige and academic careers typically depend on the ability of researchers to identify the few "essential" elements of a phenomenon. In other words, academia tends to prize the ability of individuals or research groups to characterize phenomena in general forms, and to abstract from the complexity of particular events, objects, or systems. Even when they perform very applied research tasks, researchers in non-profit institutions have fewer incentives to perfect the use of their research outcomes, as the latter are less likely to be employed for practical purposes without further developments and modifications, especially on a large scale. In short, the prestige and careers of academic scientists do not depend on whether and how their results work in reality. In contrast, in order to gain rents from their research, industrial scientists and engineers have to come up with products and processes that work satisfactorily (and not necessarily optimally) in practice, and that satisfy user needs. They cannot be content with understanding problems in general and abstract forms. They have to deal with the complexity and idiosyncra-
Science as a source of economic value
11
tic features of the particular applications of knowledge (i.e., of particular phenomena, and of related objects and systems), which are typically ignored by abstract representations. And this implies greater reliance on trial and error and experimentation. (See also Rosenbrock, 1988.) Science as a guide to search A number of studies have looked at the economic returns of scientific research. Griliches' work on hybrid corn pioneered this field (Griliches, 1957, 1958, and 1960). Research on how a cross between genetically different plants produces a plant with higher yield dates back to Mendel and Darwin. Its scientific development took place during the first thirty years of this century. Griliches found high public and private returns to R&D expenditures on hybrid corn in the USA during the period 1930-50. Following Griliches, various studies estimated the impact of basic research expenditure in firms on innovation and productivity (Mansfield, 1980 and 1981; Link, 1981, 1982, and 1985; Griliches, 1986). Using industry- and firm-level data, they found a positive and statistically significant effect of basic research on productivity growth. While this is an important result, as it establishes the existence of a positive correlation between scientific research and performance, the purely empirical nature of these works has offered little clue about the factors that link scientific research to innovation and productivity. Some authors have suggested that the contribution of scientific research to innovation can be framed in the context of search theory. Search models originated from Stigler's (1961) article about the economics of information. Nelson (1961), Evenson and Kislev (1975 and 1976), David and Stiglitz (1979), Dasgupta and Stiglitz (1980), Nelson (1982), and Nelson and Winter (1982) laid down the framework for applying search models to problems of innovation. Basic research produces information that guides applied R&D towards trajectories that are more likely to generate successful outcomes. Itfocuses the search for innovation in the downstream stages, thereby raising the productivity of applied research and development. Applied research successes are like draws of balls of a given color from urns that contain balls of different colors. Basic knowledge supplies information about the urns that are more likely to contain a larger proportion of balls of the desired color. Thus it helps the researchers to make more informed choices about the urns from which they want to draw their samples (David, Mowery, and Steinmueller, 1992). The formal specification of these models posits that innovation takes place in two stages. At the first stage, researchers study a number of
12
Introduction
techniques (urns). They then determine the yields of the techniques that were examined. At the second stage, they employ the technique with the highest yield to develop the innovation. Assume that (Xl9X2,...,XN) is the vector of yields ("values") of the N techniques that have been examined. F(X) is the distribution function of the JTs, which are independent and identically distributed (i.i.d.), and are defined on a closed interval; C is the (constant) marginal cost of experimenting with a new technique. The problem boils down to determining the optimal N; this is the N that maximizes the expression E max(Z1,X2,...9XN) — CN, where E is the expectation operator. The distribution function of Y(N) = max (Xl9X2,...9XN) can be obtained from the distribution functions F(X) of the Ts. One can then compute the expected value of Y(N). Under fairly general assumptions about F(X), E Y(N) increases with N at a diminishing rate. Hence, using a continuous approximation, the expression d(E Y)/8N= C determines the optimal N. The latter is a function of the marginal cost parameter C, and the parameters of F{X). Basic knowledge implies that innovators conduct their search within a set of more productive techniques. Suppose that F(X,0) is a family of distributions such that F(X, 0) decreases with 0 for all Ts different from the boundaries of its support. Greater 0 implies a greater probability of higher yields (first-order stochastic dominance), i.e., researchers search for the best "urn" within a "better" set of techniques. This implies a higher expected value for the best technique. Nelson (1982) showed that if dominance is additive, viz. G(X+0) = F{X), with 0, the optimal N under F() and G() does not change. If it is multiplicative, viz. G(X-0) = F(X), with 0 > 1, the optimal N increases. Along these lines, Arora and Gambardella (1994a) developed a model that distinguishes between different components of the knowledge-base of firms. The model distinguishes between ability to utilize and ability to evaluate scientific and technological information. Firms receive a signal y correlated with the true economic pay-off Zof an innovation project. Given the signal,firmsdecide whether to perform applied research and development. The expected profit-maximization problem boils down to the choice of a threshold value y* for the signal y. The firm will carry out applied research and development at cost K if and only if y>y*. Given the prior distribution of y, a higher threshold implies that, ex ante,firmsare less likely to perform a project with signal y. It was assumed that the conditional distribution of X on y depended on two parameters, y and 6. The former controls for the degree of first-order stochastic dominance. Given j , as y increases, higher values of Zoccur with greater probability. Firms are more likely to extract higher value from a given technology y, which would typically stem from better experimen-
The pharmaceutical industry
13
tation and development capabilities - "ability to utilize." The parameter 6 controls for second-order stochastic dominance (in the sense of Rothschild and Stiglitz, 1970). With greater 0, firms predict Xgiven y with lower error. They have greater "ability to evaluate" innovation projects before conducting applied research and development, i.e., they have greater ability to predict the "true" value of the innovation project. It would be natural to think that the latter ability depends on the stock of scientific knowledge available to the firm: science enables researchers to predict the outcome of experiments with greater precision. The model shows that the threshold y* is non-increasing in y. Other things being equal, greater ability to utilize implies that firms will carry out projects at lower signal levels as well. The intuition is that with better experimentation and development capabilities, a given firm can derive an acceptable pay-off even from projects with relatively "bad" priors. Ceteris paribus, with greater 0, y* is non-decreasing. Firms focus on projects with better signals. The intuition is that their predictive capability makes them more certain that good signals are associated with good pay-offs. Hence, they focus on the latter. Put differently, they can discard, with greater confidence, projects that appear to be less appealing.6 The pharmaceutical industry The drug sector is a distinctive example of the changes in industrial research discussed in the first section of this chapter. For many years industrial discoveries of new drugs have rested on empirical tests of a great many molecules in the laboratory to assess their chemical and biological activity. The approach was termed "random screening" or "molecular roulette" to emphasize that the entities to be tested were chosen on the basis of scanty a priori knowledge about their action (Schwartzman, 1976). Many economic studies and technical reports have discussed the changes that molecular biology and genetic engineering are exerting on pharmaceutical research.7 They have highlighted the expansion of technological opportunities, and the opening of largely unexplored areas of research. However, most of these studies have not really deepened the mechanisms through which these advances translate into greater research productivity. This book will argue that molecular biology and genetic engineering, together with important developments in computerized drug design and instrumentation, are offering the opportunity to extend the applicability of a "scientific" method of drug research. Industry scientists can increasingly make use of fundamental knowledge about human metabolism and the action of drugs, and they can employ advanced experimentation technologies. This enables them to select more carefully and more rationally the
14
Introduction
molecules to be tested in applied laboratory research. It is the guide-tosearch nature of science that matters here, along with the ability to perform the search more efficiently using new experimentation technologies. The book then looks at the implications of such changes for competition, strategic behavior, and market structure. The focus on upstream research is justified by the fact that innovation in the drug sector presents some important peculiarities. The generation of new drugs depends in large measure on activities that occur at the outset of the R&D process. Early research stages play a more meaningful role than in other industries, and they are the most creative steps of the drug innovation cycle. This is not to argue that other factors are unimportant. After discovery, firms perform long clinical trials to assess the safety and efficacy of new compounds before commercialization. Clinical trials command significant outlays of resources. From an economic point of view, they are a formidable barrier to entry into the research-intensive ethical drug business. But clinical trials are somewhat routine activities. They produce "yes-or-no" information about the performance of new products. In order to make effective use of information from clinical trials to improve products, companies would need to go back to basic analyses of molecular structures and laboratory tests. This implies that feedback cycles are longer and more costly than in other industries, which reduces the economic advantages of exploiting information from downstream development.8 Moreover, the tempo and organization of clinical trials are governed by fairly standardized regulatory procedures, which leaves little room for action on the part of companies. A good organization of clinical trials certainly influences the ability of firms to bring new products into the market. But they tend to determine second-order differences in the competitiveness of innovative drug firms, especially if compared with differing abilities in earlier research stages. The length of feedback cycles also explains why learning by using is less effective than in other industries. In fact, market diffusion of new drugs generates important information. Undesired effects that occur with tiny probabilities, and that are insignificant at the scale of clinical trials, may become sizeable when the drug is administered to a much larger number of patients in the market. However, like clinical trials, the market supplies dichotomous information about product performance. Unlike, say, an electronic device, drugs cannot be rapidly improved better to suit user needs. Companies would need to go back to research and long regulatory tests. Sometimes users provide information about new and different applications of drugs. This is an increasingly important phenomenon. However, as we shall see in later chapters, the ability offirmsto differentiate product applications, or to utilize information about side-effects or new
The pharmaceutical industry
15
uses of drugs from clinical trials, also depends to an important extent on their scientific expertise, and their knowledge about the basic structure of drugs and their biochemical action. Finally, in this industry, production is not a critical competitive factor. New compounds do not require extensive process modification. They are manufactured using fairly established technologies (US DoC, 1984; ADL, 1988). Moreover, the market of individual drugs, including important ones, is not very large. Unlike bulk chemical products, the size of the market does not justify conspicuous investments in chemical engineering to build big plants and attain large economies of scale. Also, new drugs that improve traditional remedies for a certain disease exhibit fairly inelastic demands. They are sold under patent, and hence under monopolistic conditions. This implies that, with constant returns to scale in production, companies have fewer incentives to pay the fixed cost to achieve a given reduction in manufacturing costs, as this would generate only modest increases in profits.9 Another important reason that justifies emphasis on upstream research is that this study focuses on a group offirmswith very similar characteristics, viz. the large US drug manufacturers. They all havefinancialcapabilities to conduct extensive clinical trials, and to meet stringent regulatory requirements. They also have large distribution assets. They are similar in many respects, but not in the extent to which, in the 1980s, they focused their strategies on the accumulation of internal scientific capabilities, and on investments in scientific research. They form a sort of natural experiment to assess the competitive potential of in-house scientific research in industry, as they allow one to abstract, in part, from other assets that are normally important for innovation-based competition. The book will address a number of questions suggested by the economic literature on science and innovation. It has already been stressed that research in the drug industry epitomizes the guide-to-search nature of science. With increasing scientific intensity of pharmaceutical research, we can also look more deeply into the incentives for rent-seeking agents to conduct research of a more fundamental character. Cohen and Levinthal (1989) and Rosenberg (1990) emphasized the use of basic research to monitor external science. Yet, one has still to produce more compelling evidence. In addition, one has to dig into the mechanisms by which in-house scientific capabilities raise the opportunities of absorbing public knowledge. Related to this is another point. If internal knowledge favors better exploitation of external findings, then, with increasing external scientific opportunities, the positions of leadership in the market are likely to strengthen. Firms with better in-house research can utilize outside knowl-
16
Introduction
edge more effectively to develop innovations. Better innovative performance will raise profits. Innovators would then enjoy a better financial position to carry out further investments in research, which would augment their opportunities of exploiting public science. Contrary to what one would expect, advances and diffusion of public knowledge may magnify, rather than diminish, differences in the ability of firms to innovate. The book will also address questions about the changes in the way companies and scientific institutions conduct their research. Among other things, it will discuss whether the need for greater scientific creativity, and the linkages with the scientific network, are altering the attitude of drug firms towards secrecy. It will also discuss whether companies tend to restrict diffusion of academic research. It has already been remarked that, for a number of reasons, research and innovation in the pharmaceutical industry exhibit important peculiarities compared with other high-tech sectors. One must thus be very careful in generalizing and extending the conclusions of this book to other industries. In no other sector are the changing nature of industrial research, the adoption of a "scientific" approach, and the related economic implications, probably as clear and apparent as in the drug sector. Nonetheless, the question which is intentionally left unanswered here is whether the pharmaceutical industry is so unique that no reasonable generalization is possible, or whether it points to some important transformation in industrial technical progress. Plan of the book The following chapter describes the changing nature of drug research and innovation. It shows how advances in genetic engineering, molecular biology, computers, and instrumentation are producing a wealth of new information which "guides" applied research towards families of molecules that are more likely to accomplish a certain therapeutic action. Chapter 3 discusses three implications of this trend. First, it examines some implications for the organization of research in pharmaceutical companies. It argues that the increasing need for scientific creativity and for absorption of public knowledge is encouraging greater openness of company research. Second, it discusses implications for the relationships between drug companies and universities. Among other things, it looks at whether more intense relationships between drug makers and universities imply greater restrictions on the diffusion of academic research. Third, it examines the growing division of labor between large drug firms and small or medium-sized research-intensive biotechnology companies. It argues that, with greater applicability of science in pharmaceutical research,
Plan of the book
17
relevant information for innovation can be cast in fairly general frameworks, which reduces the costs of transferring it across organizations. With suitable contracts and intellectual property rights, this encourages specialization and division of labor among agents with different resources and comparative advantages in the various stages of innovation. Chapter 4 develops case studies of eight large US drug manufacturers. The case studies show that, in spite of the public nature of science, firms with better in-house scientific expertise, and that organized their research in more open and "liberal" ways, obtained greater benefits from utilization of external knowledge, and they could translate it more effectively into competitive advantages. Chapters 5, 6, and 7 present three empirical studies of the economics of pharmaceutical research and innovation. Chapter 5 develops a model of drug discovery. The model is estimated using data for the fourteen largest US pharmaceutical companies between 1968 and 1991. Among other things, the empirical results show that the productivity of laboratory research in pharmaceutical companies increased considerably in the 1980s. This suggests that, apart from advances in the technology of experimentation, greater use of scientific knowledge has made the search for candidate drugs more effective. While chapter 5 looks at discovery, chapter 6 focuses on drug development and commercialization. It presents a model of innovation and competition in the industry. The model is estimated using data for the same fourteen US drug companies during the period 1968-91. The empirical results highlight a number of features about innovation, competition, and profitability in the US drug sector. A notable result is that, other things being equal, in the 1980s the productivity of the R&D expenditures of the largest drug companies increased relative to previous decades. Chapter 7 examines the propensity of drug and chemical manufacturers to develop external linkages with other parties in biotechnology. Using data on eighty-one large US, European and Japanese drug and chemical corporations, it estimates the demand for different types of co-operative linkages between these firms and other parties (mainly universities and small-medium biotech companies). The empirical results show that, from the point of view of largefirms,these different linkages represent complementary strategies. The empirical results also show thatfirmswith greater in-house capabilities in biotechnology display greater propensity to enter into external alliances. Chapter 8 concludes the book, and speculates on policy implications and future trends in the pharmaceutical industry.
Science and innovation in pharmaceutical research
Introduction: the drug innovation cycle
Drug innovation is composed of fairly standardized steps, which are designed in the main by regulatory authorities. At each step compounds are tested for particular properties or characteristics. Only a fraction of the compounds that enter a certain stage are considered viable, and they are moved to the next stage. Figure 2.1 synthesizes the typical sequence in the drug innovation process. New drugs stem either from organic chemical synthesis or from the separation of compounds produced by natural micro-organisms (Wardell, 1979; ADL, 1988; FDA, 1988). In recent years, biotechnology drugs have stemmed from the fermentation of genetically engineered micro-organisms or cell fusion (monoclonal antibodies) (OTA, 1984). There is probably no other industry wherein the impact of scientific research, whether basic research or experimental science, is as direct as in the drug sector.1 Scientific knowledge in organic chemistry, microbiology, biochemistry, physiology, pharmacology, and, more recently, molecular biology and genetic engineering, has provided the framework for drug discovery from the early decades of this century. Pre-clinical research consists of laboratory screening of molecules (bioassays and animal tests) to evaluate their therapeutic potential and toxicity. Scientists test a great many compounds before finding some that look promising for clinical trials: Sometimes, scientists are lucky andfindthe right compound quickly. More often... hundreds or even thousands must be tested. In a series of test tube experiments ... compounds are added one at a time to enzymes, cell cultures or cellular substances grown in a laboratory. The goal is to find which show some chemical effect. Some may not work well, but may hint at ways of changing the compound's chemical structure to improve its performance. The latter process alone may require testing dozens or hundreds of compounds. (FDA, 1988, p. 9.) 18
The drug innovation cycle Pre-clinical R& D
IND review (30 days)
Initial synthesis
19
Clinical R & D
NDA review
Post-marketing surveillance
Adverse reaction reporting
Phase I Phase II Phase I
Animal testing
(Short term)
Survey/ sampling testing
• •
(Long term) Average: 18 months Range: 1-3 years
Average: 5 years Range: 2 - 1 0 years
Average: 2 years Range: 2 months-7 years
Fig. 2.1. Stages in the development of new drugs. The FDA stages are shown in black. Average time from initial synthesis to NDA approval is approximately 100 months. (Source: FDA, 1988.)
Before clinical trials, in the USA firms have to apply to the Food and Drug Administration (FDA) for an Investigational New Drug Exemption (IND). The FDA can reject IND applications if there is insufficient evidence that the new compounds can be safely tested in human beings. Clinical trials comprise three stages. Going from stage I to stage III, a new compound is administered to an increasing number of patients. While stage I focuses on the toxicity of the product, stages II and III focus on its effectiveness. Typically, stage II and especially stage III involve a large number of patients, and they take place over a long period. This is to prepare an accurate profile of the drug, define its dosage, and evaluate sideeffects that occur with small probabilities or that reveal themselves only after several months. In the USA, once a new compound has passed the clinical tests, companies have to present a formal request to the FDA to market the drug - the New Drug Application (NDA). The FDA can accept or reject the application or, alternatively, require that the new medicine undergo further
20
Science and innovation in pharmaceutical research
Table 2.1. US drug development costs by stage Percentage of total development costs Discovery/applied research Pre-clinical animal testing Long-term animal studies Clinical testing phase I Clinical testing phase II Clinical testing phase III
8-24 4-5 8-10 4^-5 24-33 33-42 100
Source: USDoC(1984).
clinical tests to assess more carefully its safety and effectiveness. Once a drug is approved for marketing, it is still kept under surveillance by the firms and the FDA. Market diffusion may reveal information that was not anticipated during clinical trials. When a drug shows unanticipated sideeffects, the FDA may require that companies add warnings on the package. If side-effects or toxicity are severe, the FDA can even withdraw the product from the market. Drug innovation is a long, costly, and risky process. The development of new drugs, from applied research to commercialization, takes on average eight to nine years (Figure 2.1). If one includes the time spent on fundamental investigation of the pathology and the properties of the compounds, it may well rise to fifteen to twenty years. Moreover, development times can vary considerably across drugs. Drug development is very costly. Development costs grew considerably over the seventies and eighties. In the USA, during the early 1970s, the cost of developing one new drug (including the costs of unsuccessful projects) was estimated to be 54 million dollars (Hansen, 1979). Around the early 1980s, it was estimated to be between 70 and 90 million dollars (US DoC, 1984). In 1986, it rose to about 125 million dollars (Wiggins, 1987), and in 1990 to about 231 million dollars (Di Masi, 1991).2 Table 2.1 reports the breakdown of pharmaceutical development costs, and shows that the later stages of clinical trials represent by far the largest fraction of drug R&D. Finally, drug development is very risky.3 Recent studies indicate that only 1 out of 5,000 compounds synthesized during applied research eventually reaches the market (Halliday, Walker, and Lumley, 1992). Other estimates indicate that of 100 drugs for which INDs are submitted, about 70 complete clinical testing phase I, 33 complete phase II, and 25-30 clear
Drug discovery in industry
21
phase III (FDA, 1988). Halliday, Walker, and Lumley (1992) also found that two-thirds of the drugs that enter phase III are marketed. This suggests that attrition rates are especially severe in earlier research stages. Compounds that overcome clinical trials phase II have a relatively good chance of becoming new drugs. However, as phase III is the more costly R&D stage, one failure out of three products may still imply a considerable loss of resources. Drug discovery in industry: from chemical screening to discovery by design Chemical screening and the origins of drug discovery in industry In one of the most comprehensive economic studies of pharmaceutical innovation in the postwar period, Schwartzman (1976) argued that pharmaceutical research is largely an experimental process. Using several examples of drug innovations, he showed that, since the 1930s, large firms have specialized in extensive chemical modification of basic compounds, and they are responsible for the introduction of most new drugs in this century.4 Large firms possessed the resources to perform systematic laboratory search for new molecules, conduct clinical trials, and market new drugs on a wide scale. Moreover, the expected rents in the search for new compounds were sufficiently high to justify considerable investments in this activity. The formation of large research-based pharmaceutical companies, and their motivation to perform systematic screening of molecules were spurred by the growth of information about the properties of various classes of molecules during the early twentieth century. This was brought about by a complex blend of scientific knowledge and experimental observation. At the turn of the century, the German bacteriologist Paul Ehrlich postulated that specific sites on the surface of bacteria cells could form bonds with chemical agents (Ehrlich's affinity theory). He then observed that synthetic dyes could kill or inhibit pathogenic protozoa without affecting mammalian cells. This led, in 1910, to the discovery of Salvarsan, an arsenic compound for treating syphilis, and one of the first man-made drugs (Schwartzman, 1976; Ganellin, 1989). Ehrlich synthesized 606 compounds before finding Salvarsan. The discovery of Salvarsan, and Ehrlich's affinity theory, encouraged large chemical companies to carry out systematic examination of hundreds of synthetic chemicals to seek drugs against venereal diseases. Not surprisingly, large-scale manipulation of molecules originated in Germany, which already featured an established chemical industry, particularly in the dyestuff sector, and already had some large chemical
22
Science and innovation in pharmaceutical research
companies. Vast screening of the sulfonamide derivatives of the red azo dye led Gerhard Domagk at IG Farbenindustrie to discover, in 1931, that sulfamidochrysoidine had a strong antibacterial action in mice. Prontosil, the first really effective drug against bacterial infections, was marketed in 1935. Following Domagk's discovery, Daniel Bovet and other researchers at the Institut Pasteur in France clarified, in 1932, that the active site of the drug was not the azo dye, but sulfanilamide, a fragment metabolite of the sulfamidochrysoidine molecule. Many firms, in various countries, began screening several modifications of sulfanilamide. This encouraged development of a host of new products including hypotensives, anticonvulsants, diuretics, and antidiabetics (Schwartzman, 1976; Achilladelis, 1991). Many sulfa anti-infectives have now been replaced by antibiotics. However, Prontosil, sulfanilamide and their immediate descendant, sulfapyridine (commonly known as MB 693, and marketed in 1938), are the products that gave rise to the modern chemical-based drug industry. Since then, large-scale modification of basic chemicals, within the laboratories of large research-based corporations, has become the predominant mode of innovation in this sector. Antihistamines, penicillins, alpha and beta blockers, and many other drugs, present remarkably similar stories.5 In all such cases, scientific knowledge, experimental observations, and the systematization of these observations supplied information to assess the chemical and therapeutic properties of certain molecules. This prompted large firms to engage in extensive applied research screening to develop new drugs via manipulation of the basic compounds. The development of beta blockers is a notable example of how complex interactions between scientific knowledge and experimental observation produced information that focused laboratory screening on specific classes of molecules. In 1895, two scientists observed that a body hormone, adrenaline, injected into cats increased their blood pressure. In 1905 an English pharmacologist, Henry Dale, systematized that observation, and showed that adrenaline had different effects on different tissues. In 1947, Raymond Ahlquist clarified the action of adrenaline on different tissues, and postulated two types of adrenergic receptors: alpha and beta. This led to the search for specific compounds via modifications of the basic adrenaline molecule. The search was restricted to compounds that either stimulated or inhibited the alpha and beta receptors. Scientists developed an array of alpha and beta stimulants against asthma, and for controlling blood pressure. In the 1960s, they introduced several new drugs that blocked either the alpha or beta action of adrenaline (alpha and beta blockers) (Wardell, 1979).
Drug discovery in industry
23
The rise of discovery by design
In the 1950s and 1960s, knowledge about the properties of the compounds that could be used to synthesize new drugs was still blurred. This implied limited ability to select the substances to be tested in laboratories. More importantly, using customary techniques of organic chemistry, scientists could synthesize an immense array of chemical entities at low cost. They could generate a great many new molecules whose chemical properties, let alone linkages with physiological and pathological conditions, were unknown. As a result, not only was vast screening of chemicals the predominant approach to drug discovery in industry, but on many occasions company scientists did not even attempt to conceptualize a priori the activity of their compounds. It is important to recognize that this was a deliberate economic decision of firms. As also discussed in Arora and Gambardella (1993b and 1994b), in order to invest in fundamental knowledge, firms need to have some expectation that they can comprehend problems in generalized forms. But when problems are very complex, the costs of comprehending them in general forms can be enormous. Trial and error is then economically more advantageous, even though it yields only localized knowledge. In the postwar years, with little knowledge, and particularly with little public knowledge, about human metabolism and the action of drugs, the "private" costs of understanding problems in general terms were considerable, especially if compared with the relatively low costs of producing a great many compounds for hit-and-miss experiments. Companies then approached drug research through "blind" chemical screenings, and made little effort to understand the basic properties of their molecules. The 1960s and 1970s witnessed the growth of basic knowledge about the properties of many compounds, and about physiological and pathological conditions. This raised the economic pay-off of investments in the fundamental understanding of drugs, and gradually enhanced the applicability of a rational method of drug research. Companies could build on a more solid basis of public knowledge. They had better prospects of being able to understand organic disorders and the action of drugs at reasonable costs and in reasonable time. They then began approaching problems by looking at the general properties of molecules, and the pathology they wanted to cure. Since then, "discovery by design" has gradually gained momentum. With discovery by design, scientists use knowledge about the causes of human disorders, the properties of drug compounds, and their action in the human organism, to conceptualize the structure of an "ideal" molecule that
24
Science and innovation in pharmaceutical research
Table 2.2. Top ten pharmaceutical products in 1991 Rank
Drug
Indication
Company
Sales ($m)
Growth (percent)
1 2 3 4 5 6 7 8 9 10
Zantac Vasotec Capoten Voltaren Tenormin Adalat Tagamet Mevacor Naprosyn Ceclor
Ulcer Hypertension/CHF Hypertension/CHF Arthritis Hypertension Angina/hypertension Ulcer Hyperlipidemia Arthritis Infections
Glaxo Merck BMS C-Geigy ICI Bayer SB Merck Syntex Lilly
15,023 1,745 1,580 1,185 1,180 1,120 1,097 1,090 954 935
10.5 14.1 7.5 5.1 1.7 12.6 (2.2) 43.4 10.3 11.3
Source: FT (1992a).
is expected to restore the altered equilibrium. The ideal molecule is then given to the laboratory chemists, who search for substances whose molecular structures match as closely as possible the theoretical model.6 Clearly, discovery by design does not eliminate the need for bioassay and animal tests. Very often the ideal molecule fails in initial laboratory trials. The latter may suggest ways of improving its design. The entity then returns to bioassay and animal tests. In most cases, despite numerous iterations between theory and experiments, the project is abandoned. On a few occasions, long iterations produce a molecule that can be submitted for clinical trials. Discovery by design does not however imply that information about the properties of molecules now springs from basic research. Theoretical compounds and the knowledge about pathological processes still depend upon information flows and feedbacks between basic conjectures and experimental observation. But discovery by design raises the efficiency of the feedbacks and interactions between theory and experiments. Experiments become more informative, as observations can be interpreted using better theoretical frameworks; researchers can associate observations about different phenomena, or they can relate observations to more general classes of phenomena. In turn, this often helps in perfecting and refining theories. Tagamet, Capoten, and Mevacor, three among the ten top-selling drugs in 1991, are major examples of discovery by design (Table 2.2). Tagamet, the first H2-antagonist anti-ulcer drug, was approved for marketing in the
Drug discovery in industry
25
mid-1970s.7 Previous remedies against ulcers were based on antacid treatments, like milk of magnesia. In the event of severe complications patients underwent surgery. Sir James Black, the scientist responsible for the discovery of Tagamet, knew that ulcers are generated by over-production of gastric acid in the stomach, and that the cells responsible for secreting the acid are triggered by histamine, a chemical agent in the human body. He conceptualized a compound that blocked the acid-secreting cells by inhibiting action of the H2 histamine receptor. Although Black was looking for a compound with a specific structure and capable of a specific action in the human body, the Tagamet project took about ten years and he synthesized about 700 compounds before coming up with thefinalmolecule.8 Capoten was approved for marketing in the early 1980s. Its discovery hinged upon scientific knowledge of the chemicals in the human body that are responsible for regulating blood pressure. Renin, a chemical produced by the kidneys, releases angiotensin I, which in turn produces angiotensin II. Overproduction of angiotensin II raises blood pressure and causes hypertension. Squibb combined this information with basic pharmacological knowledge. Squibb's scientists designed atom by atom a compound that inhibited overproduction of angiotensin II, thereby blocking the rise in blood pressure (WSJ, 1987b). Merck spent decades studying how the body produces cholesterol. Early studies showed that cholesterol, a wax-like substance naturally produced by the human body, can clog the arteries that deliver blood to the heart, and cause heart attacks. During the 1970s, Merck's scientists isolated an enzyme (HMG-CoA) which is responsible for starting the production of cholesterol. This directed pharmacological research to designing a drug that could either inhibit HMG-CoA or prevent the cells from using it. In 1978, Merck's scientists isolated lovastatin, the Mevacor compound, which blocks the production of cholesterol (BW, 1987; FDA, 1988). Discovery by design has also spurred more rational approaches to the quest for diverse applications of drugs. The search for diverse uses of compounds has been common in this industry for many years. Information about new uses of drugs has typically arisen from observation of sideeffects in clinical trials or in the market. For instance, minoxidil, a drug that treats high blood pressure, also stimulated hair growth. Upjohn used this information to develop Rogaine, a hair restorer. Similarly, Naltraxone, approved against heroin addiction, proved to be active against Kaposi's sarcoma, a form of cancer associated with AIDS (FDA, 1988). But now new applications can be made less dependent on fitful and fairly casual observations. As scientists know how a certain drug acts, they can rationally plan ways of modifying the compound, and redesign it to perform different activities.
26
Science and innovation in pharmaceutical research
Aspirin, the most common medicine in the world, is a good example. It was introduced by Bayer in 1899. Since then, it has shown several therapeutic effects. Among other things, it proved to be an efficacious remedy for all sorts of pains, from simple headache to painful disturbances produced by tumors. However, the lack of basic knowledge about its action had prevented companies from developing aspirin-based medicines for many years. Only over the past twenty years have scientists gained more solid knowledge about the chemical and biological function of aspirin. Studies have shown that aspirin produces an anti-thrombogenic agent that inhibits the formation of blood clots in arteries, an important cause of heart attacks. Blood clots are caused by undesired agglomerations of blood cells, platelets, which stick together. The anti-thrombogenic agent produced by aspirin keeps platelets separate. Since the mid-1980s, doctors have prescribed aspirin in combination with heparin, another drug, to cure or prevent strokes (W. Smith, 1991; WSJ, 1992).9 Knowledge about the function of aspirin has also encouraged researchers to develop more potent medicines that prevent blood clots by hindering the agglomeration of platelets. In 1987, a group of scientists at the University of California, San Francisco, isolated the gene that contains information about the molecular structure of platelet receptors. As we shall see in the next section, knowledge of receptor structure is critical in the design of drugs that conform to receptor sites, and correct unwanted behavior of cells. Publication of the gene prompted many companies - both large companies like Merck and Smithkline, and small biotech firms like Cor Therapeutics - to invest in the development of "super-aspirins," i.e., medicines that could act on platelets more effectively than the common aspirin compound (WSJ, 1992). The spread of discovery by design Scientific advances that spur discovery by design: cell receptors and receptor technology
The applicability of discovery by design has expanded considerably during the 1980s. Like Salvarsan and the sulfa drugs, information about the properties of new compounds stems from complex interactions between scientific knowledge and experimental observations. But now theoretical tools cover a broader number of cases, and they are clarifying increasingly complex phenomena. Moreover, advanced instruments, and greater computational power, have enabled scientists to examine complicated molecular structures, and to control their chemical reactions.
The spread of discovery by design
27
Discovery by design has benefited from important developments in the field of cell receptors.10 Scientific research on receptors dates back to Ehrlich's affinity theory, and the development of beta blockers. Receptors are proteins on the surface of cells. They activate or block the functions of cells, i.e., the production of chemicals or other proteins. Human pathologies often result from the abnormal functioning of some cells: cells are inactive when they should be active or they are active when they should be inactive. Drugs bind to receptors. They can cause an inactive cell to respond (the drug is an agonist), or they can stop the undesired activity of a cell (antagonist).11 By studying the structure of receptors, scientists can design a molecule that binds to receptor sites, and activates or inhibits cells. Receptors have particular geometrical configurations, and the geometrical shape of the drug has to match the receptor like a key fits a lock. Moreover, electrical charges of drug molecules must be complementary to the atoms of receptors. Positively charged atoms of drugs must face negatively charged atoms of receptors, as two equal charges would repel each other. Drug compounds must also be stable. They must not decompose until they develop sufficiently tight bonds with the proteins. Finally, drugs must have a consistent dynamic activity. After they bind to an enzyme, they must produce chemical reactions that switch receptors on or off in desired ways. Proteins are sequences of amino acids. The genes in human chromosomes contain the codes to identify the sequence of amino acids. In turn, the sequence of amino acids contains information to determine the molecular structure of proteins. Scientists have isolated some human genes, and they have obtained information about the sequence of amino acids. Using genetic engineering techniques, they have been able to clone genes, and produce large quantities of a given protein. This has supplied sufficient material to study the proteins, and identify their molecular structure. Moreover, complex theories and experimental techniques in molecular biology have revealed the existence of a number of new receptors in the human body, and they have helped in identifying the structure of some of them. Molecular biology and genetic engineering have also given a great impetus to the possibility of predicting and controlling side-effects. Scientists have found that many receptors in different human organs have similar structures. Thus, drugs that are designed to bind to a certain receptor may also bind to similar receptors in other parts of the body, thereby causing side-effects. If scientists could devise the specific regions of the receptor molecules that differentiate otherwise similar receptors, they would be able to design drugs that fit uniquely the target receptor sites. Similarly, advances in the knowledge of cell receptors have enhanced the
28
Science and innovation in pharmaceutical research
opportunities of rational search for differentiated uses of drugs. Scientists have discovered that many new biotechnology drugs have different effects. With detailed knowledge of their action and molecular structure, researchers can select their activities. For instance, the human growth hormone was first commercialized by Eli Lilly and Genentech in 1985 to treat children affected by dwarfism. Scientists knew that dwarfism is caused by the failure of the human body to produce the growth hormone. They also knew that the production of the hormone slows down with aging. Lilly tested its hormone against some effects of age, and found that it has reinvigorating effects. The human growth hormone also helps in building lean muscle mass. Unfortunately, this has encouraged a black market among athletes. The same effect, however, has spurred research into using the hormone against obesity (NYT, 1990g; BW, 1990b).12 As discussed in the previous section, observation of the side-effects or new uses of drugs in clinical trials or in the market has been common in the pharmaceutical industry for many years. However, with limited knowledge about the properties of compounds, it was very costly just to acquire more solid evidence about the effects noted in clinical tests or in the market. Companies had to spend a great deal of time back in the laboratory, and they faced high risks that the observed effects were ephemeral. Costs and risks were not significantly lower than alternative projects "starting from scratch" or arising from other information (like evidence of effects produced by other compounds in laboratory assays). As a result, firms did not exploit systematically information gained during the development or commercialization stages. But now companies have better prospects of being able to comprehend and plan new applications of drugs at the outset of the research process. This is economically more advantageous than pursuing "casual" observations from clinical trials or the market. Firms can take advantage of the increasing returns that will naturally arise when research and experimentation about a given compound produce information that can be used for different applications. Moreover, they can select applications on a logical basis, which raises their likelihood of success. As a matter of fact, many firms are now investigating the diverse effects of their molecules (see, for instance, BW, 1990b). Relatedly, with increasing knowledge about protein receptors, and more rational approaches to drug research, we can expect greater significance for learning and first-mover advantages in specific therapeutic areas. The limits of discovery by design Although encouraging, the new advances should not disguise the immense difficulties that still persist in drug design. Scientists have found the
The spread of discovery by design
29
sequence of amino acids of only a few human proteins, and they have worked out the molecular structure of only a very small fraction of them. 13 Geometric shape, electrical charges, and, to some extent, stability pertain to the statics of drug representation. The dynamics of drugs, i.e. their chemical and biological activity inside the human body, is even more complex. Theoretical knowledge and experimental techniques in this area present ample room for improvement (CW, 1987a; NS, 1988).14 Moreover, the problems associated with similarity of receptors and multiple activities of drugs still dominate their advantages. Receptors are often so similar that it is extremely difficult to design drugs that bind to one receptor and not to others. In addition, many receptors are unknown. Hence, the likelihood of side-effects remains high. Biotechnology drugs still have too many effects, and the ability of scientists to rationally select them is very limited. These remarks simply confirm that the complexity of problems in industry can be so high that the guidance offered by fundamental knowledge is typically very incomplete. This is especially true of the pharmaceutical industry, which deals with a system, the human body, that is far more complicated than mechanical or electronic systems. There are so many and complex receptors, functions, and biochemical reactions in the human organism that drug design will rely for a long time on speculation in situations with a great deal of uncertainty. Clearly, this also implies that drug discovery will never do without long trial-and-error experiments in labs, and, of course, without clinical trials. It would nonetheless be unreasonable to deny that important progress has been made since Salvarsan and the sulfa drugs. When looking at advances in science we are often struck more by the fact that discoveries open new and wider areas of our "ignorance" than by the solution to problems that then become commonplace. Yet, it is precisely where science unravels at least part of the complexity of problems, and problems become more customary and less appealing from a scientific point of view, that industry looks for new product opportunities. The pharmaceutical industry has been no exception. For instance, in the early 1980s, scientists thought that advances in molecular biology and genetic engineering would pave the way to the development of drugs with very complex molecular structures, and capable of actions as complex as human proteins. By knowing the structure and function of human proteins, scientists believed they could replicate such structures and functions to design complex protein-based pharmaceutical products, based on "large" molecules. But in the course of the decade, it became clear that the economic usefulness of the advances in the life sciences did not rest on the possibility of developing complex protein-based drugs. More pragmatic opportunities arose from the possibility of increasing the efficiency of the
30
Science and innovation in pharmaceutical research
discovery of "traditional" chemical products. Knowledge of protein structures and their function has supplied useful models of cell receptors to which drugs can bind. This has facilitated rational design of "conventional," non-protein drugs, thereby giving a notable impetus to research in this area. Frontier scientific research in top academic institutions and other research centers still focuses on complex molecular structures, and the development of complex protein-based drugs. This is important, as scientific institutions are very active in the discovery of new receptors, and in clarifying their three-dimensional shapes. But many companies are now using the advances in the life sciences primarily to produce non-protein drugs and protein-mimetics based on small molecules. From an economic point of view, chemical drugs have many advantages over protein-based products. Their molecular structure is much simpler, which simplifies rational design. Moreover, biological drugs have complicated delivery systems. They cannot be swallowed, as pills are destroyed by enzymes in the gut and do not get into the bloodstream. Biological drugs are also much more expensive to manufacture. Finally, disputes about legal protection of natural substances have not yet been resolved, and the patenting of biological material still faces a great deal of uncertainty. 15 Serotonin drugs Many companies are developing drugs via analysis of their molecular structure and linkages with specific receptor sites using molecular biology and genetic engineering techniques. Some of these drugs are under regulatory approval for marketing, and some of them have already been commercialized.16 The family of serotonin-based compounds is a notable example of drugs developed via molecular design.17 The market of serotonin drugs is expected to tap a few billion dollars in the mid-1990s. Serotonin is a chemical secreted by the brain. It regulates a number of activities. According to where it deposits in the body, it can increase or decrease blood pressure, suppress headache, influence appetite and sexual activity, control anxiety, depression, memory, drug addiction, schizophrenia, and aggressive behavior. Serotonin was first isolated in 1948. Yet, systematic research on this molecule started only in the mid-1970s, and it was greatly spurred by advances in cell receptors. Different actions of serotonin depend on the receptors to which it binds, and therefore on the cells it activates or inhibits. For instance, activating serotonin receptors in the brain and certain blood vessels stops migraine headaches, whereas blocking them moderates anxiety and depression. The
The spread of discovery by design
31
discovery of new receptors and the visualization of their structure enables scientists to identify the cells responsible for various activities of the chemical. Researchers can then design drugs that fit the target receptors, and select different activities of the molecule.18 Research on serotonin drugs has already scored some successes. In 1986, Bristol-Myers obtained FDA approval for a serotonin-based anti-anxiety drug which is also being investigated as an anti-depressant. In 1987, Eli Lilly obtained market approval for Prozac, an anti-depressant. Prozac is the most important commercial success in this field so far. It is expected to reach one billion dollar yearly sales in the 1990s. In 1989, Ciba-Geigy obtained approval for a serotonin compound against compulsive disorders, which cause people to repeat obsessively a certain action (like handwashing). In 1990, Glaxo introduced a drug that prevents vomiting in chemotherapy patients, and it is waiting for FDA approval of a drug against migraine headaches. Glaxo's drugs are especially noteworthy, as they are opening important new research trajectories in this area. Unlike Prozac, which simply increases serotonin levels, they mimic the effects of serotonin on cells. Many other companies are developing serotonin drugs to treat disorders like cardiovascular diseases, schizophrenia, drug and alcohol addiction (BW, 1988a and 1992a; NYT, 1990b). The discovery of Prozac is another important example of drug design. Scientists showed that when serotonin, which is produced by the brain, is re-absorbed too quickly by nerve cells, it triggers depression. They realized that by slowing down the absorption of serotonin, they could improve mood. Eli Lilly's researchers rationally designed a compound that retards absorption of serotonin. Prozac was also designed in a fairly selective way to match specific receptors, and hence minimize side-effects. Reports indicate that Prozac seems to have fewer side-effects than traditional nonserotonin anti-depressants. However, clinical analysts also warn that Prozac has not yet reached the cumulative volume of sales of previous remedies, and hence unforeseen side-effects may show up in the future (NYT, 1990c). Prozac also appears to have other effects. Lilly's researchers have observed that it reduces weight, and they are studying it to treat obesity (BW, 1988a). Other serotonin drugs have shown diverse effects. Glaxo's drug that prevents vomiting in chemotherapy patients targets a receptor that is present both in the gut, where the drug acts, and in the brain. The brain receptor controls various neurological conditions. Glaxo's researchers are thus studying how to use the compound to treat anxiety and schizophrenia. Development of serotonin-based medicines still encounters several difficulties. The discovery of new serotonin receptors facilitates detection of
32
Science and innovation in pharmaceutical research
other effects. However, it also points to side-effects that, although not yet observed in patients, can be rationally anticipated. For instance, Glaxo's drug against migraine acts on receptors on blood vessels around the brain. This suggests that the drug may generate cardiovascular problems. For this reason, the FDA is delaying its approval, even though the drug has not yet shown such effects in patients. More generally, as more serotonin receptors are discovered, scientists realize that they are so similar that it will be extremely hard to design drugs that act selectively on desired targets (BW, 1992a). These difficulties suggest that a notable contribution of the new scientific discoveries is the prediction of failures rather than successes. Prediction of failures is important. Firms can save time and costs by ignoring research lines that are anticipated to be sterile. Unfortunately, the present scientific advances do not seem to be as effective in telling scientists what to do. And this implies that, in searching for what to do, companies still have to rely, to a good extent, on trial and error and somewhat "blind" experiments. Scientific instruments and computers in drug research Advances in experimentation technologies X-ray crystallography and nuclear magnetic resonance are the two spectroscopic techniques normally employed in drug research to "observe" molecular structures.19 Drug researchers have used X-ray crystallography for many years. This technique is based on the analysis of the diffraction of X-rays passing through protein crystals. The diffraction pattern reveals the spatial distribution of electrons, and the location of the atoms of the molecule. The analysis of the molecule then depends upon theoretical knowledge of how X-ray diffraction is linked to the density and spatial distribution of electrons. Relatedly, the effectiveness and precision of X-ray crystallography has benefited from advances in the science of crystals, and the development of theoretical models about atomic units and how they arrange to form molecular masses. This is important as it suggests that, although scientists try many molecular structures before finding one that satisfactorily pictures the three-dimensional shape of a protein, experimental research is strictly related to more fundamental understanding of phenomena. Theories are critical because observations of infinitesimal phenomena, like proteins, cannot be made by the naked eye (and neither can they be made by powerful microscopes). Hence, phenomena have to be deduced from theoretical relationships among objects and variables. In other words, scientists
Scientific instruments and computers
33
perform a great number of hit-and-miss experiments. But when the latter are made using advanced instruments, they depend upon complementary progress in theories. Nuclear magnetic resonance (NMR) has recently emerged as an alternative approach for inspecting molecular structures. Molecules are exposed to radio waves in magnetic fields. Atoms respond by emitting their own radio waves. The frequencies of the radio waves produced by atoms can be interpreted to devise the approximate distance between atoms. This, combined with known chemical properties of the molecule, helps in determining the position of atoms. Again, experimental observations are not made by the naked eye, but are deduced from theoretical principles.20 Mathematical analysis of how the diffraction of X-rays or the frequencies of atomic radio waves transform into three-dimensional structures of molecules, is performed by computers. Computers elaborate the data. They also visualize the protein structure on screen. In the 1950s and 1960s, researchers examined X-ray crystallography data through charts, oscilloscopes and photographs, and they built brass or plastic models of molecules. With limited computational capabilities and visualization techniques, they could analyze only very small molecules. The impressive growth of computational power in recent years has made it possible to inspect incredibly complicated compounds. This has furthered considerably the growth of discovery by design by providing models of complex cell receptors which can be used to design drugs that match given receptor sites. The productivity of X-ray crystallography and NMR has increased considerably during the past twenty years. Eberhardt et al. (1991) estimated that using today's X-ray crystallography instruments it takes one-tenth of the time, one-twenty-first of the cost, and one-tenth of the labor hours needed twenty years ago to perform the same baseline experiments. Apart from conventional X-ray crystallography, synchroton crystallography, which uses more powerful synchroton radiation sources, has developed at an even faster speed. Time reduction, cost reduction, and labor-hour reduction of synchroton crystallography have been of the order of 1:14, 1:39, and 1:15. The same ratios for NMR are, respectively, 1:9,1:7, and 1:8 (Eberhardt et al., 1991). In spite of these advances, the analysis of protein structures is still a very complicated task. In the first place, basic knowledge about how to relate Xray diffraction or frequencies of atomic radio waves to protein shapes is incomplete. As suggested earlier, computational chemists have to perform systematic trial-and-error tests to find a reasonable solution of a protein shape. They normally run different models of how a protein could be generated from crystallography or NMR data, and they produce many
34
Science and innovation in pharmaceutical research
protein structures all consistent with the given data. Long and tedious work is needed before they can come up with one protein shape from a given amino acid sequence.21 Moreover, the cost of instruments has increased with their capabilities. Eberhardt et al. (1991) estimated that, on a current-dollar basis, the capital cost of today's instruments is about the same as the capital cost of the same instrument twenty years ago. However, their life is much shorter. This implies that today's scientists have to carry out far more experiments per unit of time to achieve the same cost per hour of effective use of the instrument. Finally, scientists need to perform more complex experiments (i.e., determination of far more complex molecular structures). Thus, even though time and cost reductions have been significant when compared with the same milestone experiment twenty years ago, researchers approach problems that require longer experimentation time and greater costs compared with the 1970s. Advances in screening technologies Advances in instrumentation are also influencing the productivity of laboratory screening of compounds. The most important developments in this area are related to the growth of knowledge about cell receptors. Researchers can increasingly assess the affinity of compounds to specific receptors via laboratory assays rather than administering the substance to animals. The great advantage of in vitro over in vivo techniques is that experiments can be performed on a much larger scale, and in less time. With animal tests, scientists have to wait for the metabolization of the compound within the animal body, whilst in vitro tests provide more rapid indications of whether the compound binds to the receptor or not. (Moreover, in vitro tests are ethically more acceptable.) Novascreen, an in vitro receptor screening system developed by Nova Pharmaceuticals,(a medium-sized biotech firm), in the 1980s, enables laboratory technicians to screen about 1,200 compounds against a receptor in a week. It takes about a month to screen the same number of compounds in animals.22 Of course, receptor technology is not a complete substitute for animal tests. Test-tube experiments do not reproduce all the environmental conditions that are encountered when receptors are inside the human organism. Scientific knowledge cannot yet fully explain why, on some occasions, compounds that block or activate a receptor in vitro do not produce the same effect in vivo.23 Nonetheless, receptor screening technology is moving at a rapid pace. Some university institutions and biotechnology companies have developed very powerful techniques.24 In 1990, George Smith of Washington Univer-
Scientific instruments and computers
35
sity in St. Louis, Missouri, and three biotechnology companies, Chiron, Cetus and Affymax, made a first important breakthrough. They devised a method of creating and examining vast libraries of biological, proteinbased compounds ("peptides") to check for potential therapeutic action. The method is based upon genetic manipulation of a virus, called filamentous phage, which attacks bacteria, and makes them produce a large number of different peptides. The latter are immersed in a solution containing a target receptor. Researchers can then determine whether any of the peptides produced by the bacteria binds to the receptor. Typically, scientists cannot use the peptides they isolate as potential new drugs, because biological material, unlike chemical drugs, is destroyed by enzymes in the gut. However, they can use the peptides as models to design chemical drugs. The main limitation of this method is that it can test only biological material, i.e. material produced by live organisms such as viruses and bacteria. This implies that the method can only produce drug models based upon the twenty naturally occurring amino acids. This rules out the countless chemical compounds that can be produced synthetically. Two biotech companies, Houghten Pharmaceuticals and Selectide, have developed alternative receptor screening methods that can test vast libraries of chemical compounds against specific cell receptors. Both methods hinge upon the following idea. Human proteins (enzymes, receptors) are composed of chains of twenty different amino acids. Drugs are approximately the size of a six-amino-acid chain. Six-amino-acid chains, called hexamers, are normally sufficiently small to sneak into bigger molecules, like human enzymes, and influence the way they work. Organic chemists have created more than 150 synthetic amino acids. They can then produce billions of different hexamers by simply combining the natural and synthetic amino acids six at a time. Houghten and Selectide have devised methods to test quickly the reaction of a huge number of hexamers against a given receptor. For instance, it may take just one week to test all 64 million possible hexamers that can be obtained by combining the twenty natural amino acids six at a time. The two methods consist of immersing millions of hexamers in a solution containing a target receptor. Radioactive material makes the reacting hexamers turn a different color. They can be isolated, and their amino acid sequence can be identified. The sequence can then be used as a model to design drugs. Many other companies, which include, apart from biotech firms, software companies, are trying to develop even quicker peptide screening methods. A few of them are trying to automate the screening process by using robots that work twenty-four hours a day and seven days a week to
36
Science and innovation in pharmaceutical research
test peptides against an array of different receptors. This is likely to have important effects on future opportunities for drug discovery. High-speed and very efficient screening methods could provide vast libraries of drug models which are known to react against certain receptors. These would be like libraries of "prototypes" which could be used as "ideal" compounds to design new drugs. Efficient screening methods may give a new impetus to extensive trial and error in drug research. In fact, as suggested earlier, the new methods arose primarily from a basic understanding of the function and structure of cell receptors. Moreover, as hexamers are not candidate drugs themselves, researchers still have to perform chemical drug designs, which rely upon rather fundamental knowledge about the properties of drugs and their action inside the human body. More importantly, some analysts have suggested that a major limitation of the new screening methods is that researchers have little ability to extract useful information from such huge libraries of peptides {ECON, 1991). Greater scientific understanding is needed in order to utilize more selectively the enormous number of peptides that can be generated in this way. Relatedly, even though researchers can efficiently screen a great many hexamers against various receptors, it would be practically impossible, even with the most sophisticated technologies, to perform blind screening of peptides against the massive number of receptors in the human body. Researchers need some guidance about the pathology they want to attack. They also need guidance to restrict the number of receptors to be examined. Put differently, efficient screening technologies will be more valuable when applied to areas where basic knowledge offers an indication of potential opportunities. Once again, it is the complementarity, rather than substitution, between basic understanding and improved experimentation technologies that drives the prospects for successful research, and for commercially useful innovations. Computers: the fundamental tools of drug design Computers have become critical instruments for drug research. We have already seen that, armed with impressive computational power, scientists can now approach the complicated mathematical equations that are necessary to work out the molecular structure of very complex human enzymes from X-ray crystallography or NMR data. Sometimes, computational chemists attempt to determine protein structures without using information from X-ray crystallography or NMR. Purely computational models rely solely on basic theories of quantum chemistry about the density of electrons and the position of atoms. However, fully theoretical models
Scientific instruments and computers
37
are typically forced on researchers by the difficulties of collecting data from experimental observations. Researchers then simplify equations using arbitrary assumptions. At present, the most effective (and feasible) models are those that use "realistic" information from X-ray crystallography and NMR to approximate immensely complicated quantum chemistry equations (NS, 1988; Science, 1992).25 Computers also convert the solution of protein models into threedimensional images. Three-dimensional visualization is of the utmost importance for comprehending the intricacies of molecular shapes. In the past, molecules were represented using balls, sticks and wires. Physical visualization presented various problems. For instance, the physical structures of complex molecules composed of thousands of atoms, such as proteins, could collapse under their own weight. With computers, scientists can easily rotate images, and observe them from different angles. Computerized images can also be very helpful in detecting regions of receptor structures where drugs can bind. Computers are also used to design drugs that match target receptors.26 The drug industry has used computers for this purpose since at least the end of the 1970s, and it was one of the first industries to employ computers to design molecular structures (Science, 1992).27 At present, most applications are static representations. Typically, given the topography of a receptor, researchers design drugs whose geometrical configuration matches suitable spaces of the enzyme. Apart from rotating the image, and observing it from different angles, they can easily add or eliminate atoms to experiment with different models of drugs. Moreover, computers perform various calculations about atomic charges, size, and other aspects of the chemical interaction between drug molecules and receptors. Static representations can rarely predict potentially effective drugs with great precision. An important reason is that they ignore dynamic aspects about the action of compounds inside the enzyme. However, they can be very useful, as researchers can eliminate many drug candidates simply because their geometrical configuration or atomic charges are inconsistent with the target receptor. Cost savings can be substantial as companies can exclude from animal tests and clinical trials a number of compounds that will not bind to specific enzymes. Dynamic simulations are still in their infancy. But some applications already offer the possibility of simulating the actions of drugs that occur in less than a trillionth of a second. The computer reproduces the action in much slower motion. Researchers can carefully study it. They can stop the process, and focus on the critical moments of the reaction. This may suggest ways of improving the structure and activity of the compound. Drug candidates never move from the computer screen directly to IND
38
Science and innovation in pharmaceutical research
application and clinical trials. Geometrical conformity to target receptors is the area where researchers are most confident about the predictive power of their computerized models. However, scientists are largely unable to account for all the environmental factors that influence the dynamic interactions between potential drugs and human enzymes. For instance, a major problem is that the polarity charges of "ideal" drug compounds tend to change when actually introduced in the live organism, and these changes are often impossible to predict. As a result, drug molecules that seem to work satisfactorily in computers often fail in animal tests (BTEC, 1990). Because of these difficulties, drug designers supply applied chemists with many alternative drug models. Applied chemists synthesize these models, and test them in bioassays and animals. New compounds for clinical tests typically arise from systematic trial and error, and information flows and feedbacks between the design stage and applied laboratory research. Indeed, the most important contribution of computers in drug research is not that they reduce the need for laboratory experiments. Instead, they enhance the efficiency of the interactions between drug design and applied research. They raise the efficiency of the entire process of designing the compound, synthesizing it in the laboratory, and turning back to the design stage to modify it according to information from the experiments. Abbott's development of a renin inhibitor exemplifies the role of computers in drug discovery. In the search for a new drug, Abbott scientists had already developed an angiotensinogen protein-drug that could bind to the target human protein, renin. They had the amino acid sequence of the protein, but lacked a three-dimensional structure of renin. The threedimensional structure of renin was important to identify sites where the new drug could bind. Laboratory chemists had already synthesized a number of molecules as potential renin inhibitors. From the computer visualization of the renin structure, computational chemists could discard two-thirds of the molecules synthesized by the applied chemists, as their structural configuration did not fit the renin receptors. Further development could be confined to only one-third of the compounds, with implied time and cost advantages. Animal tests showed that the molecule could not be given orally. Enzymes in the gut reduced the entity to its inactive form, which was not revealed by computer models. The molecule went back to the computer screen. Scientists realized that chymotrypsin, an enzyme in the gut, was responsible for making the drug inactive. Given the molecular structure and the receptor sites of chymotrypsin, they redesigned the drug compound so that it was no longer affected by the enzyme. When the redesigned compound was given to pharmacologists to prepare it for administration, the substance proved to be insoluble. The
Are we really moving forward?
39
compound returned to the computer. The computational team tried to identify groups of atoms that, on the one hand, could be modified without affecting the activity of the drug, and, on the other, could improve its solubility. They found a residue that was far from the active site of the molecule, and could be eliminated without affecting the therapeutic properties of the drug. The elimination of the residue considerably improved the solubility of the substance. The newly redesigned molecule passed the animal tests, and it was moved to clinical trials {BTEC, 1990). Conclusions: Are we really moving forward?
The present trends in pharmaceutical research present a serious puzzle. Progress in the life sciences and experimentation technologies has not translated into indisputable increases in the number of discoveries and commercializable innovations. A comparison with electronics is striking. In electronics, since the end of the Second World War, advances in basic chemistry and physics, and in the technology of research, have brought about a significant succession of innovations. The miniaturization of chips and the continuous introduction of new generations of computers are only the most apparent examples. In the pharmaceutical industry, there is still a great deal of uncertainty about whether massive and socially very costly developments in theoretical and experimental sciences are producing sizeable economic and social advantages. This chapter has highlighted the difficulties of drug design. More importantly, the number of new drugs commercialized worldwide between 1987 and 1991 has declined (Figure 2.2). Several reasonable explanations can be advanced. Major developments in biological sciences and experimentation technology have occurred only since the 1980s. Commercializable innovations will come up in the 1990s or even in the new century. Relatedly, the human body is an incredibly complex machine. Enormous efforts in science unravel only tiny portions of this complexity. The "easy" research veins have already been exhausted, so now greater research endeavor is necessary to produce fewer discoveries. Finally, drug development costs and the stringency of regulation are rising. Fewer pharmaceutical companies invest in innovation, which reduces the number of new drugs (Thomas, 1987; Di Masi et al.9 1991). The discussion in this chapter suggests another possible explanation. The present growth in scientific understanding has had a greater impact on our ability to predict failures rather than successful drugs. The discovery of new receptors reveals side-effects. With computerized analyses of drug geometry, researchers can immediately discard compounds that will not bind to target enzymes, and thus will be ineffective. Previously, with limited
Science and innovation in pharmaceutical research
40 70
63 60-
55 48
50T3
i.
44 38
403020-
10 -
0
1987
1988
1989
1990
1991
Fig. 2.2. Number of new drugs commercialized worldwide, 1987-91. {Source: Prous, 1992.)
ability to predict failures, side-effects or inefficacy would be observed only after long clinical trials and more often only after the drug had been in the market for some years. Thus, more drugs were tested and commercialized only to discover later on that they were ineffective (or worst, toxic). But now many ineffective drugs, or drugs with anticipated side-effects, can be discarded at the outset of the research process, and hence fewer drugs will be tested and commercialized. The desirable facet of this trend is that the fewer drugs that will be tested and sold will have fewer side-effects, and they will be more likely to be really effective drugs.28 In effect, attrition rates and the extent of laboratory experimentation do not seem to have diminished. Halliday, Walker, and Lumley (1992) found that in the 1980s only one substance out of 5-6,000 synthesized was marketed. Cox and Styles (1979) obtained similar results using 1970 data. In the 1960s, Vane (1964) obtained an even higher yield (1:3,000). Some of the earlier remarks apply here as well. The benefits of the 1980s' advances will materialize in the 1990s, and scientists approach increasingly complex problems, which require more extensive trial and error. The idea of a diminishing number of pre-clinical trials rests upon the assumption that the predictive power of science reduces the need for experimentation. Other things being equal, this is true, as the outcome of some experiments can be logically anticipated. As also discussed in chapter 1, in Arora and Gambardella (1994a) it was formally shown that when profit-maximizing agents have greater ability to predict the future pay-offs of innovation projects, they focus on fewer projects with higher expected
Are we really moving forward?
41
value. The number of applied research trials shrinks, as less projects are carried out. (However, expected experimentation yields increase, as companies develop only projects with greater expected pay-offs.) But other factors may work in the opposite direction. As extensively discussed in this chapter, the efficiency of trial and error has risen (because of computers, new laboratory screening techniques, etc.). This encourages companies to perform more hit-and-miss experiments. The efficiency of the interactions between upstream theorizing and downstream experimentation has also increased. Complementarity has become more pronounced, which could be more than offseting the substitution effect. Indeed, the search models discussed in chapter 1 suggest that advances in science enable companies to search for innovations within a set of more productive "techniques." This implies that the optimal number of techniques tested is non-decreasing (Nelson, 1982). Put simply, search theory indicates that, with advances in science, experimentation becomes a more productive input. Hence, rational economic agents will not "buy" less of it. Moreover, if measured by the number of drugs marketed per compound synthesized, as in the studies above, yields will not necessarily increase. The critical contribution of present advances in the life sciences and experimentation technologies lies in the possibility of generating higher quality drugs, i.e. truly effective drugs, and drugs with fewer side-effects. This does not imply more drugs per compound tested. More careful analyses ought to look at the value of marketed drugs per compound tested, assuming that the market value of drugs reflects greater quality in the product. In sum, in comparing the social and economic benefits of the present scientific and technological advances with their massive social costs, the focus should be on whether drug researchers can increasingly eliminate lower quality drugs at the outset of the research process. Benefits will arise if more effective drugs are commercialized, and drugs have fewer side-effects. This analysis, however, is beyond the scope and the possibilities of this book. One would need much finer data, and to wait for a few years for present scientific and technological progress to have a sizeable impact on commercializable products. Nonetheless, what is important for us, as economists, is that economic agents, viz. pharmaceutical firms, and the scientific community, seem to have good expectations that the current advances will lead somewhere. They are actively investing in molecular biology, genetic engineering, computerized drug design, and new experimentation technologies. Irrespective of whether such things will work, this should be enough to raise the interest of our profession. Expectations are sometimes all that matters in our discipline. Not only can they be self-fulfilling, but more importantly they influence economic behavior. And it is to this that we now turn.
3
Economic implications of greater scientific intensity in drug research
Scientific research in pharmaceutical companies Growing incentives for scientific research in drug companies The growing scientific intensity of drug research has increased the incentives for pharmaceutical firms to invest in basic understanding of their compounds. As argued in the previous chapter, the decision of firms to invest in fundamental knowledge is an economic one. When problems are very complex, basic scientific and technological research cannot produce concrete product opportunities in reasonable time and at reasonable cost. Trial and error is economically more advantageous. This is why in the 1950s drug companies engaged in systematic screening of compounds, with limited attempts to understand the action of drugs. But the growth of basic biological and pharmacological knowledge, and the increased efficiency of experimentation, have raised the possibility of comprehending relatively complex problems in economically useful ways. Especially from the 1980s, firms have had greater expectation that knowledge of the properties and action of compounds could be used to disclose new product opportunities. Advances in the life sciences, instrumentation, and computers have enhanced the efficiency of information flows and feedbacks between theoretical knowledge and experimentation. It is now easier and less costly to perfect and correct theories (and theoretical designs of drugs) using advanced instrumentation techniques. At the same time, knowledge of drugs and the human body supplies valuable frameworks to utilize experimental observations more coherently. As a result, pharmaceutical firms increasingly use theoretical tools and new research technologies to predict the therapeutic effects of their molecules before synthesizing them in the laboratory. Clearly, greater economic incentives to invest in understanding the properties of drugs are reinforced by the externalities produced by the 42
Scientific research in pharmaceutical companies
43
growth of public knowledge. Firms can focus on problems that have already been clarified, in large part, by public discoveries, and hence that have greater potential for economic returns. Typically, scientific institutions isolate genes and identify protein structures. Problems in these areas are scientifically the most intriguing, and the pay-off to scientists (reputation, status, etc.) is the greatest. But these areas are remote from tangible product opportunities. Hence, companies are unlikely to specialize in them. However, as our examples of super-aspirins and serotonin in the previous chapter demonstrated, as soon as scientific institutions isolate new genes or identify new receptor structures, pharmaceutical companies rush to exploit the new discoveries. Sometimes they provide supplementary contributions of a fundamental character. They clarify the function of some receptors, they find other functions, or they discover similar receptors. More often, they use knowledge about newly discovered genes or receptors to design new drugs, elucidate their action, etc. Their reliance upon external knowledge also implies that pharmaceutical companies invest in upstream research as a means of monitoring public scientific advances (Cohen and Levinthal, 1989; Rosenberg, 1990). Many large drug corporations own divisions that perform rather basic research. Apart from studying problems such as the isolation of genes, the determination of protein structures, or improvements thereof, they perform very general studies of drug compounds, or they study pathologies on a very general basis. Such activities can be extremely useful from a commercial point of view. If a company discovers, say, a certain protein structure, it may gain significant first-mover advantages. More importantly, by performing basic research, company scientists remain aware of developments in their own disciplines; by keeping their research skills alive, they have greater ability to utilize new findings and translate them into commercially useful innovations. Apart from technological opportunities, pharmaceutical research exhibits solid appropriability conditions, which fact augments the incentives of drug companies to invest in understanding the basic properties of their molecules. We saw that the fundamental input of drug discovery is not science per se, but a complex blend of scientific knowledge and experimental research. As we shall see in more detail in the next section, both generation of scientific knowledge and mastery in experimental research depend a great deal upon tacit skills, learning, and individual and teambased talents and expertise. More generally, they depend on organizationand team-specific resources, which cannot be easily transferred to other teams, let alone to other organizations. In addition, unlike many other industries, patents are an effective means
44
Economic implications
of protecting drug innovations (Taylor and Silberston, 1973; Mansfield, Schwartz, and Wagner, 1981; Mansfield, 1986; Levin et al.9 1987; Cohen and Levin, 1989).1 An important reason is that pharmaceutical innovations take well-defined forms, to wit new compounds. Innovations can thus be described, and hence patented, more easily. This also makes it easier to identify the object of patents, which reinforces legal protection. Because patents are effective, pharmaceutical companies rush to patent their new entities as early as possible. They patent their molecules after preclinical research. Thus, not only are patents an effective means of protecting drug innovations, but they are also a means of protecting the output of basic and applied research. This is an important factor in raising the incentives of firms to invest in upstream research. Even though scientific findings diffuse in the external environment, companies can appropriate their economic applications, and hence the principal source of rents produced by investments in fundamental knowledge. Patents also disclose the content of innovation. Yet, drugs are complex molecules when compared for instance with bulk chemical products. "Inventing around" from information in patents is more difficult than in other branches of the chemical sector. Also, patents are granted a few years after application. This can confer important lead-time advantages. The innovator can start animal and clinical tests as soon as the innovation is generated, i.e. months before the information is disclosed by publication of the patent.2 If compounds enter clinical trials earlier than competitors, they are more likely to reach the market sooner. First-comer drugs, especially if they are important advances over existing remedies, can secure high customer loyalty (hospitals and doctors), which makes it more difficult for second- or third-comers to gain significant market shares.3 Organization of research and scientific creativity
The growing use of scientific knowledge in drug discovery has important implications for the organization of company research. When new compounds resulted primarily from extensive screening of chemical entities, the critical resource for drug discovery was sufficiently large laboratories. Only large laboratories could reach the threshold size of trials that was needed to attain one or a few potential new products. Dynamic economies of scale were also important. Over time, large laboratories could accumulate information about the properties of many families of compounds by testing them under different conditions (e.g., for different pathologies).4 Moreover, discovery was more effectively managed by hierarchical organizations. With limited applicability of science, large-scale screening of compounds hinged upon fairly routine tasks. Firms had to employ many
Scientific research in pharmaceutical companies
45
"technicians" who had to perform rather conventional experiments to observe and record how different molecules reacted. Typically, hierarchical structures are more efficient for managing large organizations that have to carry out fairly repetitive activities. With greater need for scientific creativity, the scale of laboratories is no longer the only critical asset for discovery. As argued by many authors, flexible and informal organizations, even relatively small in size, are often conducive to the production of ideas of great originality (e.g., Arrow, 1983; Sylos Labini, 1992). This is especially true of pharmaceutical research (FT, 1989b; Shelley, 1991; Angard, 1991). As a result, pharmaceutical firms have to pay increasing attention to organizing their research so as to stimulate ingenuity. This also intensifies the differences in the character and organization of the research and development stages in this industry. Drug development is still based upon long and routine clinical trials, which are more effectively managed by large hierarchical structures with considerable financial and organizational capabilities. In contrast, the far greater need for creativity implies that in discovery important changes have to take place. (See also Angard, 1991.)5 Successful drug companies will increasingly organize their discovery process around a group of talented scientists. The personality of great individuals has always played an important role in pharmaceutical innovation. Many drugs are associated with the names of the head scientist who managed the project.6 Sapienza (1987) examines the story of Tagamet, and the roles of Sir James Black and William Duncan, the scientists responsible for its discovery. As she points out, successful completion of the project depended not only on their scientific expertise, but also on their leadership skills, and their ability in creating an atmosphere of great intellectual fervor in the lab. Moreover, in order to encourage creativity, top managers will have to loosen hierarchical control on company scientists. Top managers have broader vision about the market, the position of their company with respect to its competitors, and the fields in which innovations have greater economic potential. Hence, they should define the key therapeutic areas wherein company researchers have to focus. But within these areas scientists should be accorded great autonomy in choosing specific research lines. They have greater technical competence to pinpoint the best opportunities for technological and even commercial success. In addition, scientists should be granted substantial independence in organizing their research and the activities of their teams. Of course, some control is necessary. It is then important to devise forms of control that avoid hierarchical pressure, which thwarts creativity. One possibility is to replicate the peer evaluation system in academia. Firms
46
Economic implications
could encourage discussion and periodic revisions of projects, through internal seminars or in other forms, among scientists and researchers working on different company projects. Companies could also encourage discussion with academic scientists. Decisions about whether to abandon a research project or to enlarge its budget should be taken by a council of senior company scientists or by a committee composed of the head scientists of all company projects. Discussions and seminars could be extended to marketing people and other company managers. They may give important suggestions to make projects closer to commercial opportunities. Conflicts may arise as marketing directors and other managers endorse business-oriented objectives too strongly, whereas scientists want to linger on the more theoretical facets of projects. An important function of top managers will then be to resolve such conflicts by making choices that preserve commercial goals without frustrating research ingenuity. Research openness in pharmaceutical firms
As they organize research around a group of senior scientists with minimal hierarchical control,firmswill also be forced to accept greater openness of their research. This issue was discussed more generally in Arora and Gambardella (1992). Basic research and computerized design of drugs imply that companies will have to employ more people with strong academic backgrounds (Ph.Ds., young researchers, etc.). These people will have internalized to a greater extent the values of the scientific community. Some of them will even remain motivated to do research just for the sake of expanding the stock of knowledge. They will be inclined to publish, participate in conferences, and communicate with their peers. A related explanation is that qualified scientists and researchers will have individual incentives to enhance public recognition of their human capital. They may want to preserve opportunities of returning to academia, or they may want to maintain visibility for possible employment by otherfirms,or they may simply want to keep a credible threat of leaving their present companies. While firms could impose restrictions to avoid diffusion of internal information, this would mean applying pressure and control. We have already argued that this may dampen creativity and frustrate motivation for research.7 More importantly, there are considerable economic benefits arising from the fact that company scientists publish, participate in conferences, and communicate with their peers. We noted the importance of in-house scientific research as a means of monitoring and absorbing external scientific knowledge. But the ability of company scientists to comprehend
Scientific research in pharmaceutical companies
47
and utilize outside knowledge will be greater if they behave like members of the scientific community (which entails publications, participation in conferences, etc.). Again, company scientists will be able to assimilate external scientific knowledge more effectively if they perform similar research themselves (including research of a more academic nature). For instance, they will have greater ability to understand and utilize public discoveries about newly isolated genes or receptor structures to design new drugs. Even though they may not discover the new genes or receptors themselves, research expertise in this field will help in better understanding issues such as the geometrical combination of drugs and receptors and their dynamic interactions. Moreover, academic scientists will be more inclined to exchange ideas with people that they deem to be part of the same "club" (Arora and Gambardella, 1992). Exchange with academic scientists is important for various reasons. Apart from clues about new research ideas, company scientists may know about discoveries before publication, which often follows discoveries only after a time-lag. Even a few months' lead in knowing about a certain finding can be very important in this industry. Also, publications, or other reports, typically do not convey complete information about a discovery. There is a great deal of useful information that can be transferred between scientists by word of mouth. The drawback of openness is that firms lose control of internal information. However, the ability to produce - or even to "re-produce" scientific knowledge depends to a large extent on organizational routines, as well as on talent and skills (including experimental skills) embodied in the human capital of individual researchers (De Solla Price, 1984). Thus, to take advantage of knowledge and information (or even compounds) generated by the research group of the firm, competitors would need to have similar "tacit" capabilities. This also implies that firms will increasingly compete for the services of eminent scientists and qualified researchers. This will be done through pecuniary and non-pecuniary rewards. Among the latter, a fairly important one will be the creation of a stimulating (and "open") research environment - not least because this would enable scientists and researchers to preserve or even enhance public recognition of their human capital, with implied personal advantages discussed earlier. Moreover, there are learning effects. Company scientists may have accumulated over time great expertise in a specific area. Even if internal information circulated outside the firm, competitors would be unable to exploit it effectively if they had not specialized early enough in the same field. First-mover advantages in promising areas of scientific inquiry, and in new therapeutic and commercial niches, can then become critical. But the
48
Economic implications
ability of firms to move quickly into promising new areas of research will depend, once again, on their proficiency in monitoring the wealth of external information, and in capturing opportunities that are created largely outside of their organization. As the earlier discussion pointed out, this ability will be greater in companies that encourage external linkages and research openness. In sum, lower control of internal information is an important drawback of openness. However, this is to be weighed against the increasing disadvantages of secrecy. The most important one is that the company is not part of the scientific network, and it cannot trigger effective links with external information. Moreover, as discussed in the earlier subsection, patents are an effective means of appropriating drug innovations. In principle, if the firm patented, earlier than its competitors, the economic output of its scientific ideas, it would not be as compelling to protect the ideas themselves. Paradoxically, stronger patent protection of drug compounds might lead to greater openness of company research. Companies would be able to secure more effectively the ultimate source of their economic rents, and the advantages of openness would outweigh even more strongly its disadvantages. Relationships between pharmaceutical companies and scientific institutions Licenses and research collaborations between drug companies and scientific institutions
Collaborations and other relationships between pharmaceutical firms and universities or other research institutions have been common in the drug industry since the early years of this century (NSB, 1982; Swann, 1988; Madison, 1989). The growth of scientific intensity in drug research will strengthen the propensity of companies, and possibly of research institutions, to enter into such relationships (OTA, 1984,1988 and 1991; Kenney, 1986). Table 3.1 lists the most important linkages between the largest US pharmaceutical companies and universities or other institutions in the 1980s. As the table shows, such relationships primarily take two forms: licenses to companies arising from independent research conducted by research institutions, and research agreements. Licensing is a common means of technology transfer from scientific institutions to drugfirms.Companies obtain rights (which can be exclusive or non-exclusive) to develop and commercialize an earlier discovery of an academic or government laboratory. The diffusion of university-industry licensing in the pharmaceutical industry is encouraged by the relative ease
Table 3.1. Agreements and other relationships (in biopharmaceuticals) between large US pharmaceutical firms and universities or other research institutions in the 1980s Firms
University or other research center
Products
Type of relation
Abbott Labs.
Nat. Inst. of Health
HIV-1 ELISA diagnostic
Abbott Labs.
University of Chicago
rDNA lung surfactant
Licensing agreement R&D agreement
American Cyanamid
Columbia University
Protein chemistry facility
$250,000 for establishment
1985
American Cyanamid (through Lederle)
Columbia University
Polio vaccine program
R&D agreement
1989
American Home Products
Stanford University
Basic cloning procedures
Non-exclusive license
1980
American Home Products
Columbia University
Patented rDNA extraction
Non-exclusive license
1984
American Home Products
National Technical Information Service
Rotavirus vaccine
Limited exclusive license
1987
Bristol-Myers
University of Alabama (with Schering-Plough)
Anti-gp 1200 against HIV
5-year development agreement
1987
Bristol-Myers
Walter and Eliza Hall Inst. of Medical Rsrch. (Melbourne, Australia)
Blood cell growth, patterns for leukemia treatment
Research agreement
1987
Bristol-Myers
MIT (Dr. Susumu Tonegawa)
Genetic operation of immune system
Cancer research study award
1987
Bristol-Myers
National Technical Information Service
Monoclonal antibodies to Chlamydia
Exclusive marketing agreement
1987
Year — —
Table 3.1. (cont.) Firms
University or other research center
Products
Type of relation
Year
Bristol-Myers
Max Planck Institute
Antiviral nucleosides
Licensing agreement
1987
Bristol-Myers
Yale University
R&D agreement on anti-cancer medicines
Renewal of 1982 agreement; BM supplies $600,000 per year for 5 years; BM has option on licenses of new products discovered by participating Yale faculty during the period of the agreement
1987
Bristol-Myers
US Dept. of Health and Human Services
DDA, DDI nucleosides for AIDS
BM to develop
1988
Bristol-Myers-Squibb
Institut Pasteur
BMS's US patent for HIV-2 and its components
Licensing agreement
1989
Joint R&D Non-exclusive license Long-run research done at Cambridge research lab located on MIT campus $70 million 10-year development agreement
1990 — —
Bristol-Myers-Squibb
Nat. Cancer Institute
Taxol
Johnson & Johnson
Columbia University
Patented rDNA extraction
Johnson & Johnson
MIT
Diagnostics and immunology
Johnson & Johnson
Scripps Clinic
Pharmaceutical products
Lilly
Columbia University
Patented rDNA extraction
Non-exclusive license
—
Lilly
Scripps Clinic
Monoclonal antibodies for us in adenocarcinoma
R&D agreement
—
1986
Lilly
MIT
Polymer-based continuous release drug delivery system Beta-adreneregic receptor
R&D agreement Research agreement
1986
Merck
Duke University
Merck
Purdue University
Monoclonal-antibodies-based nasal spray for rhinovirus infections
Support R&D of Michael Rossmann
1985
Merck
Mass. Gen. Hospital (with Biogen)
rDNA Mullerian inhibiting substance (MIS)
Development and marketing agreement; M to fund clinicals and marketing
1987
Merck
Istituto Gentili
MK-217 for osteoporosis
Licensing agreement
1989
Merck
Institut Merieux
Merck
Instituto National de Biodiversidad (Costa Rica)
R&D agreement Vaccine for hepatitis B, diphtheria, tetanus, and polio Natural sources of new drugs in R&D agreement; M funds res. rain forests of Costa Rica in exchange for exclusive rights on products
Monsanto
California Institute of Technology
Instrument design
Development agreement
Monsanto
Columbia University
Patented rDNA extraction
Non-exclusive license
Monsanto
Shemyakin Institute (Moscow)
Biological research
3-year research agreement; M funds research in exchange of rights to market discoveries in the West
1989
Monsanto
Washington University (St. Louis)
Proteins and peptides, virusresistant plants
4-year research agreement which extends previous 8-year agreement, bringing total funding to nearly $100m; M provides additional $9m per year in 1991-94
1990
—
1991 1991
— —
Table 3.1. (cont.) Firms
University or other research center
Schering-Plough
Products
Type of relation
Mass. Gen. Hospital
Monoclonal-antibodies-based lytic agents
Research agreement
Year —
Schering-Plough
Oregon State University
rDNA vaccine for bovines
Funding research
—
Schering-Plough
Pennsylvania State University
rDNA technology
Industrial affiliation program
—
Schering-Plough
Scripps Clinic
rDNA alpha interferon for basal cell carcinoma
Product development and Royalty agreement
1986
clinical trials
S'kline
Ohio State University
Feline leukemia virus vaccine
S'kline
Walter Reed Armv Medical Center
Malaria vaccine
S'kline
Pasteur Vaccines
S'kline
National Institute of Health
AIDS therapeutics
R&D grant (to NIH)
1987
S'kline
Stanford University
Processes and products
$7.8 million to construct a building for Center for Molecular and Genetic Medicine; SK has right of first refusal
1987
S'kline
University of Cambridge
SK provides $2.25 million to Cardiovascular and autoimmune diseases, and virus establish a research institute in return for options on patents infections and commercialization of resulting inventions
1987
Exclusive rights for malaria vaccine to SK rDNA surface antigen hepatitis Worldwide licensing agreement B vaccine
1980 1985 1986
S'kline
Johns Hopkins University
SK provides $2.2 million for Diagnosis and treatment of allergic and respiratory diseases 5-year research in return for rights of first refusal
S'kline
Washington Research Foundation (with Genentech)
Patented technology for recomb. protein in yeast
License to SK
1988
Squibb
Oxford University
Diagnostic and therapeutic agents for treatment of central nervous system disorders
S provides £20 million for a neuroscience facility and projects in exchange for patent rights on products
1987
Sterling Drugs
Harwell Labs. (UK government)
Macrosorb chromatography system
Licensing agreement
1984
Sterling Drugs
Purdue University
Antiviral agents
Research agreement
1987
Sterling Drugs
Memorial Sloan-Kettering and Columbia University
Research agreement
1991
Syntex
Stanford University
Polar-planar differentiation agents Molecular genetics
Syntex
Hong Kong Insitute of Biotechnology
Drug screening research
Joint venture
Upjohn
California Institute of Technology
Instrument design
Development agreement
—
Upjohn
US Dept. of Commerce
Licensing agreement
—
Upjohn
Battelle Memorial Institute
rDNA procedures to turn vaccine virus into live vector Human vaccines
Research
1985
Upjohn
University of Kansas
Drug delivery and transport system
Joint collaborative research
1986
1987
Syntex provides $1.5 million to 1987 Stanford's Center for Molecular Genetics and Medicine 1991
Table 3.1. (cont.) Firms
University or other research center
Products
Type of relation
Year
Upjohn
University of Goteborg (Sweden)
Compounds for treatment of central nervous system disorders
Long-term R&D agreement
1987
Upjohn
Centre Paul de Broca de rinserm, University College, London
Compounds for treatment of central nervous system disorders
U has exclusive rights from collaboration
1987
Upjohn
Stanford University
Sleep research
Research
1987
Upjohn
Nat. Cancer Institute
Tetraplatin
Licensing agreement
1989
Warner Lambert
University of Auckland (New Zealand)
Anti-cancer
Joint development of new anticancer drug
1983
Note: Table covers relationships found in sources below for all US-based pharmaceutical firms among the first twenty companies in terms of 1986 US drug sales. Pfizer not included because no relationship with research institutions was found in sources below. Monsanto is included because, even though not among the first twenty companies in terms of 1986 US drug sales, it is investing significantly in the pharmaceutical business. Years of some agreements are not given in source. Source: Bioscan (various years) and Predicasts Index (various years).
Companies and scientific institutions
55
with which one can transfer the output of basic and applied research in this field. Licensing is a more effective means of technology transfer when the subject of the license can be meaningfully described in blueprints (Mowery, 1988; see also Arora, 1991). University research often leads to the synthesis of new compounds. This can be a deliberate goal of academic projects, or a by-product of more fundamental research. As compounds are well-defined objects, licensing exhibits low transaction costs. For instance, because the object of licensing is well defined, and hence is easily understood by both parties, information transfer displays fewer risks of opportunistic behavior due to asymmetric information or other reasons (Mowery, 1988). Moreover, companies can readily utilize the license, without incurring the costs associated with the transfer of substantial complementary tacit information. As scientific institutions do not normally have resources (and incentives) to carry out long and costly clinical trials (and to commercialize products), they yield the rights on their discoveries to gain funds for future research. While licensing arises from research conducted independently by scientific institutions, research agreements entail industry funding of university research projects. As shown by table 3.1, they are typically extended arrangements covering 3-5 years or even longer periods. On some occasions, they are pure research contracts wherein companies fund research carried out by university staff. Firms benefit from contractual rights on outcomes, and they can take advantage of easier access to faculty and university outputs (preview of papers, participation in seminars, discussions with faculty, etc.). Agreements can also take the form of genuine research collaborations. Companies fund research jointly conducted by university and corporate researchers. 8 Firms can gain important benefits from broad research collaborations with academia. For instance, Monsanto's executives declared that the 1982 four-year research agreement with Washington University (St. Louis) proved to be extremely successful in supplying the company with research and product opportunities. This justified the renewal of the agreement in 1986 and in 1990. Moreover, they pointed out that the 1982 agreement justified the 1985 acquisition of Searle, which established Monsanto's entry in the drug business (BW, 1986). Why do pharmaceutical firms finance university research?
The propensity of large pharmaceutical companies to finance academic research can be ascribed, again, to a favorable combination of technological opportunity and appropriability conditions. Opportunities arise because of the "short" distance between scientific research and commercia-
56
Economic implications
lizable applications. We saw that scientific institutions specialize in activities such as isolation of genes and determination of protein structures which can be very useful for drug design and development. Moreover, universities often conduct research down to the discovery of specific compounds, and on many occasions they perform initial development stages (animal tests or even initial tests on human beings). In addition, large pharmaceutical firms can spread the costs and outcomes of "generic" research conducted by their university partner on a fairly wide range of product categories in which they have commercial stakes. More generally, the scientific community is responsible for most of the present advances in molecular biology and genetic engineering, and it is the repository of most of the knowledge-base in these fields. Thus, apart from specific product opportunities, research collaborations with academia help companies in acquiring familiarity with the "new" science-base of this industry. In view of the short distance between science and commercially useful innovations, prompt acquisition of such capabilities is critical. Relatedly, research collaborations can give rise to privileged relationships with specific institutions. This puts companies in a better position vis-a-vis their rivals in exploiting the future discoveries of their partner. 9 Good appropriability conditions reinforce the incentives of pharmaceutical companies to invest in linkages with academia. In many cases, university-industry contracts involve options or rights on academic discovery (table 3.1). Hence, appropriability can be an inherent feature of such contracts. The effectiveness of patents in protecting pharmaceutical innovations additionally helps in securing the economic outcomes of linkages with universities. In working with academia, companies gain knowledge that can be used for in-house drug discovery. They can then appropriate, through patents, the outcomes of the knowledge acquired from collaborations with research institutions. This encourages investments in academic research, in spite of the fact that part of the knowledge produced by company-sponsored projects (and that represents the basis of some of its innovations) may diffuse in the outside environment. Appropriability also arises from the fact that companies can establish preferential channels with the institutions with which they collaborate. They can secure research information that is not as easily and promptly available to rivals. Tacit abilities are critical in scientific investigation (De Solla Price, 1984). Through collaborations, companies accumulate experience about methods and other subtleties of the research conducted by their partners, which can be acquired only through systematic, "day-to-day" relationships (e.g., expertise in using instruments such as X-ray crystallography, NMR, or computers for drug design). The relatively short distance between academic research and commercia-
Companies and scientific institutions
57
lizable innovations also implies that, with prompt access to findings, companies can pre-empt rivals in developing innovations within specific therapeutic fields. They can start laboratory research, as well as animal and clinical tests, earlier. Because the timing of clinical trials is in large part determined by regulatory conditions, there is little chance that rivals that entered downstream development a few years or even a few months later can reach commercialization before the first innovator. The importance of lead-time advantages in the early stages of pharmaceutical research thus implies that relationships with scientific institutions can bestow a dynamic competitive edge despite the fact that academic findings, including those accomplished through company-sponsored research, eventually become, at least in part, publicly available. Research collaborations and restrictions to academic freedom Appropriability of academic research and closer and more intense relationships between pharmaceutical firms and scientific institutions, raise an important issue. Companies may attempt to secure greater appropriability of the university research that they support by imposing greater restrictions on disclosure. Clearly, this problem will intensify as scientific research gains more "economic" value, and it becomes more directly and intimately connected with profitable opportunities.10 It will become even more intense if public research funds shrink. Academia will have to resort more systematically to firms for research money. Also, with reduced outside opportunities, universities will be in a weaker bargaining position to negotiate restrictions on diffusion with companies. The extent to which companies seek greater restraints on disclosure in joint research programs with academia is controversial. In discussing the 1987 alliance between Squibb and Oxford University, D. Smith (1991) noted that the major reason that prompted Oxford University to collaborate with Squibb was public budget research cuts in Britain during the mid1980s. However, he also noted that the agreement posed minimal restrictions on academic freedom, and it has produced net benefits from an academic perspective. For example, Oxford University scientists have been able to produce many valuable publications, and they have patented a few new discoveries. Similarly, Gluck (1987) examined university-industry relationships in biotechnology using data from over 1,200 questionnaires sent to faculty members in forty US schools. In comparing faculty members with and without industry support, he found that the former had better publication records, and spent more time on genuine academic activities (e.g., time with students) (see also Blumenthal et al, 1986).n In contrast, Cohen, Florida, and Goe (1992) examined 500 university-
58
Economic implications
industry research centers in a number of disciplines in the USA, and found that industry placed restrictions on academic freedom in 57 percent of the cases. Moreover, they found that industry-supported university research tended to shift the focus of academia from basic subjects to more applied research. Also, table 3.1 shows that, among the most important universityindustry agreements in the US pharmaceutical industry during the 1980s, there are no consortia, and practically all relationships involve only one firm. It has already been remarked that most linkages are relatively longterm research contracts, and in many of them companies have exclusive rights or first options on discoveries. Direct appropriability of university findings then appears to be an important factor in encouraging pharmaceutical companies to link with academic institutions. Kenney (1986) discussed at some length whether long-term contracts between large pharmaceutical corporations and universities alter academic freedom and research autonomy. He presented case studies of a number of important university-industry collaborations in biotechnology during the 1970s and the early 1980s. His case studies covered agreements between very large firms and fairly large groups within universities (e.g., entire departments, or substantial parts thereof), which have lasted for a number of years and have involved considerable sums of money (among others, Harvard-Monsanto, Massachusetts General Hospital-Hoechst, HarvardDu Pont, Washington University-Monsanto). Kenney expressed concern about restrictions on universityfindingsbeing made available to the public. In the cases that he examined, corporations tended to direct and control the research agenda of universities. He expressed his concern even though, as he suggested, both the corporations and the universities minimized the extent to which the relationships had compromised academic freedom. His greatest source of concern was the extent to which companies exercised a generalized control on the activities of scientists. In most of his agreements, the companies exerted control on public presentations (e.g., in professional meetings or seminars). Clearly, control was confined to research related to company-sponsored programs. But if corporate funding covers a substantial share of the research time of a certain department, or of individual scientists, this can be a considerable fraction of the overall research work of an institution or scientist. Moreover, with extended corporate funding, and given the natural overlap among the various research projects of scientists or university departments, it can be very difficult to distinguish between non-sponsored research (which can be freely disseminated) and sponsored work. Companies exerted significant control on other scientists' activities as well. Kenney reported the story of a scientist who joined Massachusetts
Companies and scientific institutions
59
General Hospital after the latter had signed a long-term research agreement with Hoechst. The new department member had a previous consulting relationship with Genetics Institute. He was then asked to pass to Hoechst any information acquired while working with Genetics Institute in exchange for similar information gathered in the department and arising from the agreement with Hoechst.12 More generally, Kenney pointed out that, in such long-term agreements, companies are really buying more than a window on technology. They are buying research assets, in the form of significant shares of university research time. Thoughtful discussion of the positive implications of, and the policy prescriptions arising from, such a complicated, and still controversial topic is beyond the scope of this book. An obvious positive implication is that greater scientific intensity of drug research will encourage tighter and more systematic interactions with research institutions, and this, in turn, will intensify debate (and controversy) about academic freedom. This will be a particularly hot topic in the pharmaceutical industry, wherein companies enjoy direct innovation opportunities from academic linkages, and hence have greater incentives to control academic knowledge. My predictions, however, are less ominous than Kenney's (see especially Kenney, 1986, p. 72). In long-term research agreements, and especially if both parties enjoy significant advantages from a successful relationship, breakdown of the relationship can constitute a serious threat for both parties. Both parties will then try to avert the failure of the relationship. They will make serious efforts to comprehend each other's needs and behavioral codes, and they will gradually build up confidence and trust. More importantly, the discussion earlier in this chapter about the advantages of open communication in company research can be extended to relationships with academia. Companies will recognize that they can derive greater benefits from linkages where scientific creativity is not frustrated by undue restraints on academic freedom, and scientists can take full advantage of open communication with the scientific network to assist their partners more productively. Interestingly enough, it is the ability of large pharmaceutical companies to appropriate the downstream outcomes of scientific research (through patents, lead-time advantages, or development and commercialization assets) that will discourage them from restraining open communication too strongly. If they are confident about their ability to secure part of the downstream outcomes of university research, they will not restrain information exchange too seriously, as they can obtain research inputs of greater value. In this connection, Blumenthal et al. (1986) found that small to medium-sized biotech firms seek stronger restrictions from their academic partners than larger firms. They have less
60
Economic implications
ability to protect upstream findings by means of downstream assets, and hence they are more concerned about direct appropriability of scientific results. Clearly, firms will impose confidentiality on research and information that have critical commercial value. Academia will certainly be more constrained than without industry connection. Yet, this does not necessarily imply that less scientific knowledge will be produced and disclosed. To the extent that companies become less obsessed with secrecy, financial support and other stimuli from industry (e.g., the use of costly instruments, novel research questions, interactions with company scientists) could induce greater production and disclosure of scientific knowledge than if no relationship existed. Naturally, these remarks are by no means conclusive. Even though companies may become less restrictive about disclosure, there are other unresolved issues. For one, would companies shift the focus of academia from fundamental investigation towards more business-oriented research? A related concern is that, as shown by table 3.1, pharmaceutical companies appear to derive greater benefits from one-to-one relationships with academia than from consortia or other relationships with several partners and this is clearly connected to the fact that they will encourage open research only if they can appropriate downstream outcomes. A certain company can then monopolize commercial exploitation of the discoveries of its university partner, despite the fact that, from a social point of view, it may not be the most efficient agent to develop them, or at least some of them. An even more serious concern is that a company can prevent other firms from utilizing discoveries that it does not utilize. Policy prescriptions are even more intriguing. Again, the critical issue here is that, unlike most other industries, pharmaceutical companies have very few incentives to establish connections with university institutions if such linkages also involve some of its rivals. This implies that university policies aimed at preserving academic autonomy from industry cannot be similar across industries and scientific disciplines. Unlike industries wherein the relationships between science and commercializable innovations are less direct, in the pharmaceutical sector university policies that discourage connections with only one industry partner could discourage all universityindustry linkages.13 Financial problems of pharmacology or biology departments could become especially severe if this occurred in periods when public research budgets were undergoing serious cuts. A sensible university policy is probably to avoid interventions that deter linkages with individual industry partners. One would have to rely on the fact that pharmaceutical companies will increasingly recognize that academic freedom and research openness is beneficial from their point of view as well. An
Division of labor
61
even superior policy, at the national level, would be to ensure that university departments enjoy sufficient public research funds, as a form of "outside option," to be in a better bargaining position when negotiating restrictions on disclosure in university-industry collaborations. Division of labor and the formation of networks in pharmaceutical innovation The rise of biotechnology firms: from integration to networks of innovators The market structure of the pharmaceutical industry is undergoing a considerable transformation. For many years, the research-intensive ethical drug segment of the pharmaceutical industry had been composed of large integrated firms, which internalized activities from research to distribution.14 Between the mid-1970s and the mid-1980s, more than 300 small to medium-sized research-intensive biotechnology firms were founded (Pisano, Shan, and Teece, 1988; OTA, 1988), and quite a number have been formed lately as well (Burrill, 1989; OTA, 1991).15 The growth of biotech firms has hinged upon the new opportunities opened by molecular biology and genetic engineering. Most biotech companies have specialized in research, and a large fraction of their sales is composed of research contracts for larger companies (Burrill, 1988). Molecular biology and genetic engineering have displaced traditional "chemical" capabilities for drug discovery. Chemical capabilities were "sunk" in the corporate tradition and "culture" of large pharmaceutical firms. Empirical studies by Henderson and Clark (1990) and Henderson (1993) showed very clearly that "old" technological routines are embedded in the organization of established firms. When radical innovations, or radically new research methods appear, organizational rigidity and inertia hinder the ability of established firms to take advantage of the new opportunities. New entrants, or newly established firms, with no sunk costs and organizational biases towards the old technology, can be far more effective than incumbents in exploiting the new fields.16 The rise of biotech firms is consistent with this pattern. However, while molecular biology and genetic engineering have had a notable influence on the discovery process, they have not really changed the type of assets that are needed for drug development and commercialization. The latter still require conspicuous financial, managerial, and organizational resources to conduct long and costly clinical trials, and extended distribution networks to sell the new products. These assets are firmly owned by established corporations, and biotech companies have encountered serious difficulties in acquiring them. A few of the largest biotech companies have discovered
62
Economic implications
important new products, which they have commercialized independently or in collaboration with larger firms (Burrill and Lee, 1991). Yet, in spite of repeated attempts to integrate forward, the vast majority of biotech companies have been unable to obtain thefinancialresources and marketing assets for independent development and commercialization of their discoveries. As a result, rather than new entrants ousting incumbent firms from the market, interesting changes have occurred in the organization of the innovation process in this industry. Many new drugs, and especially those that hinge upon the new advances in molecular biology and genetic engineering, no longer result from activities that are integrated within large corporations. They stem from network-like arrangements between biotech companies and large drug manufacturers, based on research collaborations, joint ventures, and the like (see, among others, Pisano, Shan, and Teece, 1988; Burrill and Lee, 1991; OTA, 1991).17 Biotech companies supply ideas, compounds, therapies, and applied research outcomes. Large companies supply complementary research capabilities (such as those that require lumpy research assets, and hence cannot be borne by smaller firms), as well as resources for large-scale development and marketing. Tables 3.23.5 show the complex structures of the external linkages of four representative drug corporations (two American and two European) - Merck, Pfizer, Ciba-Geigy, and Hoffmann La Roche. Networks of similar complexity could be drawn for practically all the largest US and European drug corporations. (See Bioscan, 1992.) Division of labor in innovation
Economists have long debated about the most effective size of innovating firms. The received Schumpeterian view is consistent with the rise of large research-based pharmaceutical corporations since the 1950s: large firms have long-term perspectives for R&D investment plans, and they have adequate resources and organizational capabilities for large-scale R&D. In contrast, Jewkes, Sawers, and Stillerman (1958) argued that most "inventions" arise from individuals or small groups. Mueller (1962) resolved some aspects of this debate. He presented case studies of Du Pont's twenty-five most important innovations between 1920 and 1950. He showed that the largest fraction of such innovations originated outside Du Pont, from smaller firms or individuals. He thus concluded that "Du Pont has been more successful in making product and process improvements than in discovering new products" (Mueller, 1962, p. 344). Mueller also pointed out that Du Pont's ability to generate major innovations did not rise proportionally with its research expenditures. He
Table 3.2. Ciba-Geigy's network of agreements Name
Products
Content of agreement
Date
Affymax N.V.
Affymax's drug discovery technology to treat diseases such as cancer, arthritis, and autoimmune disorders
R&D agreement - Ciba-Geigy will develop and market any compounds discovered, Affymax will receive milestone payments and royalties on sales
7/91
Agricultural Genetics Co. Ltd. Agricultural Genetics Co. Ltd.
Microbial insecticides RFLPs
R&D agreement Contract R&D agreement
6/85 1/89
Agri-Diagnostics
MAb diagnostic kit to detect fungal disease affecting field crops and horticultural specialties
R&D agreement
3/86
ALZA
Transderm® drug delivery system for scopolamine, and nitroglycerine
Marketing agreement
ALZA
OTC cough/cold/allergy products based on ALZA's OROS® technology
Marketing agreement through Ciba Consumer Pharmaceuticals
10/8
Aphton Corp.
Aphton's immune system stimulants against parasites in animals
Development agreement
2/92
Applied Microbiology Inc.
AMBICIN antimicrobial peptides to treat bovine mastitis
Worldwide licensing agreement
3/91
Biogen
Ciba-Geigy's yeast promoter system for vaccine production
Licensing agreement
1/86
Biosys
Biopesticides for turf and ornamental pests
Exclusive distribution agreement for the USA
4/91
Table 3.2. (cont.) Name
Products
Content of agreement
Date
Biosys
Nematode strains as bioinsecticides
4/92
Calgene Chiron
Disease resistance in plants IGF-I and II for osteoporosis and bone and muscle healing (somatomedin)
Chiron
rDNA vaccines for AIDS, hepatitis A, hepatitis non-A, non-B, CMV, herpes, malaria EMIA diagnostic tests, including test for thyroxin levels
R&D and worldwide marketing agreement Research agreement Development and production agreement, C-G has option for worldwide licensing rights 50/50 joint venture called The Biocine Co. Corning holds license to develop, manufacture, and market worldwide through Ciba Corning Development agreement - Genencor to research, develop, and manufacture, C-G to market the products worldwide $42m. worldwide licensing agreement to develop, manufacture, and market
2/84
Five-year, $30m. co-operative research agreement - C-G to have worldwide exclusive marketing rights
11/90
Collaborative Research Inc. Genencor International Inc.
Enzymes for the pulp and paper industry
Genentech
Animal health care products including animal interferons and other lymphokines Antisense technology
ISIS Pharmaceuticals
6/86 6/86 10/86
6/92 7/85
North Carolina Biotechnology Center, R.J. Reynolds
Plant biotechnology
Funding for Fellowship Program of Consortium for Research and Education in Plant Molecular Biology
Noven Pharmaceuticals
Noven's transdermal estrogen delivery system
Exclusive licensing agreement for USA and Canada
11/91
Oregon Health Sciences University
Nucleic acid, protein, and peptide chemistry, and molecular biology
$2.5m., five-year co-operative agreement to promote the exchange of information
1987
Panlabs®, International
Natural products discovery
Multiyear research agreement
1991
Plantorgan (Germany)
Eglin, Hirudin
Collaboration and licensing agreement
Tanox Biosystems
Protective MAbs against HIV
Collaborative development agreement
1989
Tanox Biosystems
MAb products for treating certain allergies
Joint development agreement - C-G to pay Tanox an undisclosed amount of money, with additional payments as product development progresses
5/90
Source: Bioscan (1992).
Table 3.3. Hoffmann La Roche's network of agreements Name
Products
Content of agreement
Date
Ajinomoto Co. Inc., Immunex
IL-2 technology
11/84
Alpha 1 Biomedicals Inc.
Thymosins
Alpha 1 Biomedicals Inc. Amgen
Thymosin alpha-1 Neupogen® G-CSF
Angenics
Screening tests
Biogen N.V. Biogen N.V.
Hepatitis diagnostic antigen B-cell growth factors
Boehringer Ingelheim Vetmedica G.m.b.H.
Antimicrobial substance aditoprim
Chiron Corp. Chiron Corp.
IL-2 patents Products based on ras oncogene research
Exclusive license outside Japan and Asia (royalties from Hoffmann via prior agreement) Hoffmann has partial buy-back rights, holds foreign rights Licensing agreement Development and marketing agreement for Europe Licensing and sponsored research agreement Supply agreement Licensing agreement, Biogen retains comarketing rights Joint development agreement, synthesized by Roche, future marketing by Boehringer Cross-licensing agreement Five-year R&D agreement, Hoffmann to share in funding of oncogene research at Chiron in exchange for exclusive marketing rights to the resulting products
1/89 9/88 7/86
11/87 11/88 12/88 1989
Chiron Corp.
IL-2 products
Joint development and marketing agreement - both companies to market Proleukin® and Roferon-A® in Switzerland and in all EEC countries except Denmark and Greece
4/90
Cortecs Ltd.
Oral delivery formulation of Roche's genetically engineered alpha-IFN, Roferon-A®
Collaborative R&D agreement - Cortecs to develop drug delivery system
8/90
Dainippon Pharmaceutical Co.
IL-1 alpha
Cross-licensing agreement
2/90
Genentech Inc.
IFN (alpha, beta)
Genentech Inc.
DNase to treat cystic fibrosis and chronic bronchitis
Development and exclusive marketing agreement Ten-year development and marketing agreement
Genica Pharmaceuticals Corp.
PCR DNA analysis
Development agreement
HA as drug delivery vehicle
Research agreement, Genzyme provides therapeutic grade HA
1987
Treatments for immunological diseases
Five-year $10m. research agreement, Roche to receive rights to license patents
11/89
Collaboration agreement to research, develop (Immunex), and market (Hoffmann) - originally signed (1/84), extended
8/86
Genzyme Harvard Univ. Medical School, Institute for Chemistry Immunex, Ajinomoto
IL-2
3/92
Immunomedics
Radiolabeled MAbs for CEA for cancer diagnostics
Licensing agreement
6/86
Interferon Sciences Inc.
Alferon® N alpha-IFN for treatment of genital warts
Marketing agreement
3/88
Table 3.3. (cont.) Name
Products
Content of agreement
Interferon Sciences Inc.
Alferon® N Injection
Licensing agreement
3/91
Marion Merrell Dow
Ro-40-5967 third-generation calcium channel blocker
Joint development and marketing agreement
4/91
MetPath Inc.
DNA probe assays for cancer, infectious diseases, and genetic disorders using PCR technology
Development agreement
7/91
Protein Design Labs.
PDL anti-Tac MAb to prevent organ rejection
Licensing agreement
1989
SangStat Medical Corp.
Pregnancy test kits to detect the presence of human chorionic gonadotrophin
Supply agreement through Produits Roche
12/90
Schering-Plough Corp.
Alpha-IFN
Cross-licensing agreement (Biogen holds European patent, Hoffmann holds US, developed through Genentech), Europe not included in agreement
5/85
Scios Inc.
Nazdel® delivery system with Hoffmann's anti-obesity and growth hormone products
Scios to conduct pre-clinical investigation
6/86
Summa Medical Corp.
MAb for imaging blood clots
Licensing agreement
10/87
Syntex Corp.
Toradol IM (ketorolac tromethamine) injectable analgesic
Joint development and marketing agreement - Syntex to manufacture and distribute, Roche to co-ordinate marketing
4/90
Date
Xenova
Novel biochemical assay for immunosuppressant screening
Roche granted exclusive worldwide rights to develop and market active compounds arising from earlier research collaboration
4/92
XOMA Corp
MAb cell lines for STD diagnostics
Licensed to Hoffmann
1985
Source: Bioscan (1992).
Table 3.4. Merck's network of agreements Name
Products
Content of agreement
Date
AB Astra AB Astra AB Astra ALZA Corp.
Tonocard to treat cardiac arrhythmia Losec for ulcers Plendil® entry for blocker hypertension Oral controlled-release form of diltiazem
1982 1986
ALZA Corp.
Sustained-release bolus delivery system for ivermectin, a bovine antiparasitic Vaccines rDNA Mullerian inhibiting substance (MIS)
Marketing agreement Marketing agreement for USA Marketing agreement for USA Licensing agreement - Merck to market drug worldwide under a royalty-bearing license, ALZA retains some rights to market in USA R&D agreement
Behringwerke Biogen N.V., Mass. General Hospital Celltech Group p.I.e. Chiron
Duke Univ. E.I. Du Pont
Oral arthritis therapeutic based on enzyme research Recombivax® HB yeast-based hepatitis B vaccine Beta-adrenergic receptor Angiotensin II (All) receptor antagonists for treating high blood pressure and heart disease
Marketing rights in Germany Development and marketing agreement Merck to fund clinicals and market worldwide, Biogen to manufacture R&D agreement Exclusive worldwide licensing agreement giving Merck right to produce and market Research agreement Collaborative research and marketing agreement - Du Pont to have exclusive marketing rights in North America to two of Merck's products (Sinemet and Vaseretic®)
3/90
1985
2/87 1989 10/84
9/89
Hoechst A.G.
Hepatitis B vaccine
ImmuLogic Pharmaceutical Corp.
Vaccines to prevent autoimmune diseases, and AIDS vaccine program
Immunetech Pharmaceuticals
Certain future products, excluding pentigetide
Distribution agreement through Behringwerke Collaborative development agreement (10/89), extended for two additional years Option to market in Germany, Austria, Switzerland, and Benelux countries
MAb M340
Licensing agreement
Prodiax® for diabetes
Licensing agreement
10/86
Lisinopril (Prinivil®)
Licensing agreement
1986
R&D agreement
6/91
R&D agreement - Merck will fund research in exchange for exclusive rights to any products developed Licensing agreement 50/50 joint venture called Johnson & Johnson Merck Consumer Pharmaceuticals Co. Acquired licensing rights to technology Development and marketing agreement Five-year research agreement
9/91
Imperial Cancer Research Technology Imperial Chemical Industries (ICI) Imperial Chemical Industries (ICI) Institut Merieux
Combined vaccine for hepatitis B, diphtheria, tetanus, and polio Natural sources of new therapeutics from the rain forests of Costa Rica
Instituto National de Biodiversidad (INBio) (Costa Rica) Istituto Gentili Johnson & Johnson
MK-217 for treating osteoporosis OTC medicines
Medlmmune, Inc. Medlmmune, Inc. Panlabs®, International
Technology to produce cellular immunity HIV MAbs Fermentation strain improvement
1986 4/92 2/84
11/89 3/89 1/91 12/91 1989
Table 3.4. (cont.) Name
Products
Content of agreement
Date
People's Republic of China
Hepatitis B vaccine
11/89
Purdue Univ.
MAb-based nasal spray for treating rhinovirus infections AIDS polypeptide subunit vaccines
Licensing agreement for vaccine production using Merck's genetically engineered materials Support R&D of Michael Rossmann
Repligen Corp.
Shionogi & Co. Ltd. Singapore Biotechnology Singapore Biotechnology SmithKline Beecham Biologicals S.A.
Recombivax® HB hepatitis B vaccine (developed by Chiron) Heptavax-B® conventional hepatitis B vaccine rDNA hepatitis B vaccine Hepatitis B surface antigens
Syva
Polyclonal and MAb diagnostics
Vical Inc.
Gene-based vaccines for infectious disease
Source: Bioscan (1992).
R&D and marketing agreement - Merck has worldwide rights to vaccine technology, Repligen to receive royalties and manufacturing rights Joint development agreement for clinical trials and marketing in Japan Six-year agreement to produce Distribution agreement Worldwide sublicensing agreement covering the manufacture and sale of recombinant hepatitis B vaccines Formed Syva-Merck joint venture to market R&D and licensing agreement
late 1985 5/87
1984
7/86 4/90 1982 6/91
Table 3.5. Pfizer's network of agreements Name
Products
Content of agreement
Date
Advanced Polymer Systems Inc.
APS microsponge technology
Marketing agreement
4/90
ALZA
OROS® Procardia® nifedipine delivery system
Joint development agreement
1985
Bayer A.G.
Nifedipine GITS
7/88
Celltech
Celltech's patented rDNA chymosin
Exclusive marketing rights for Bayer overseas Licensing agreement
Collaborative Research, Dow Chemical Co.
Rennin for making cheese by genetic manipulation of yeast
Research, licensing, and worldwide marketing agreement - Dow assigned marketing rights to Pfizer
1/88
1988
Ecogen Inc.
Bt-based biopesticides
Manufacturing agreement
1/92
Genzyme
Orthopedic products using hyaluronic acid technology
R&D agreement through Howmedica
2/87
Ligand Pharmaceuticals Inc.
Therapeutics for osteoporosis and other bone diseases
Five-year joint discovery program
5/91
The Liposome Co.
TLC D-99
Licensing agreement - Pfizer to have worldwide marketing rights
11/90
Microvascular Systems, MPS (IGI) Moleculon
Liposome formulations for injectable animal vaccines
Development agreement
9/87
Poroplastic® controlled-release disc used in Paratect® antiparasitic bolus system
Manufacturing agreement
Table 3.5. (cont.) Name
Products
Content of agreement
Date
Natural Product Sciences Inc.
Central nervous system agent with excitatory amino acid antagonists
10/90
Neurogen Corp.
Drugs to treat anxiety
Oncogene Science
Anti-oncogenes (tumor suppressor genes), oncogenes, TIF, TGIs (tumor growth inhibitors) as cancer therapeutics
Petroferm Scios Inc.
Petroferm's Emulsan® Products to treat diabetes and obesity
Three-year agreement and licensing agreement with $2.75m. Pfizer investment (1987), original agreement extended for two years Five-year R&D collaboration - Pfizer to fund Neurogen's R&D program in exchange for worldwide manufacturing and marketing rights to one anxiolytic drug developed through the collaboration Five-year, $12m. collaborative R&D agreement, licensing options, Pfizer to fund construction of lab and to have exclusive worldwide rights to all resulting products (7/86) - agreement extended to add five years, $16m. in research support to Oncogene for a new project based on tumor suppressor genes Manufacturing and marketing agreement Five-year, $30m. joint development agreement through Metabolic Biosystems Inc. (unit formed by Scios for venture) - Pfizer and the new unit will share manufacturing and marketing
2/92
12/90
1984 1988
Wellcome Group XOMA Source: Bioscan (1992).
Pfizer's beta thymidine for Wellcome's Retrovir® Xomen E5® MAb-based therapeutic for septic shock
rights for certain therapeutics in the USA and Canada, Pfizer will hold exclusive manufacturing and marketing rights for the rest of the world Long-term supply agreement Development and marketing agreement
6/87
76
Economic implications
speculated that Du Pont's research expenditures during the period 1920-50 would be more highly correlated with its less spectacular, incremental improvements: Although ... [large industrial firms]... may be perfect vehicles of applied research and innovation, they may not have adequate economic incentives for sponsoring the ideal environment for conducting the basic research leading to the inventions underlying important innovations, or they may not be able to create that environment. (Mueller, 1962, pp. 345-6.)18 Many authors have now recognized that large firms have comparative advantages in large-scale development and commercialization of innovations, whereas smaller groups are better suited for upstream research (e.g., Arrow, 1983; Holmstrom, 1989; Scherer and Ross, 1990; Sylos Labini, 1992). Big firms have large organizations, which are critical for systematic product and process development, and they have extended commercialization assets. They also face lower capital costs. A reason for this is that they can resort more extensively to internal funds. Small firms cannot finance large-scale development projects internally, and they have to borrow. In innovation, moral hazard is severe, and capital markets command a premium. Moreover, large firms face a lower cost of external capital, as they can spread uncertainty over a larger number of activities (and innovation projects), and more generally their solvency is less at risk. But the flexible and informal organizations of small groups facilitate invention and the production of ideas. Arrow (1983) suggested that the organizational "distance" between inventors and the people responsible for internal financing of innovation is greater in large firms than in small ones. This generates greater information loss in internal communication. Larger firms thus face more intense problems of asymmetric information, and therefore smaller firms make closer to optimal investments in more novel and riskier innovation projects (provided that they can finance them). (See also Holmstrom, 1989.) This suggests that a division of labor in innovation can be socially efficient, as it allows different agents (small and large firms) to focus on the tasks in which they have comparative advantages (Arora and Gambardella, 1993b). Arrow (1983) also argued that large and small firms will recognize their "natural" abilities, and they will specialize accordingly. Large firms will recognize that they can save budgets by resorting to small innovative firms for ideas and risky innovation projects.19 Small firms will then invest in research even though they know that they will be unable to develop and commercialize their inventions, as they expect to sell their research outcomes to larger firms. Patents and intellectual property rights will be critical to prevent opportunistic behavior in information exchange. But with adequate intellectual property laws, a market for research outcomes arises:
Division of labor
77
The existence of markets for research outcomes ... alters the incentives for research within large firms . . . For now the firm has an alternative supply of research outcomes on which to base its development of innovations. The constraints on its development expenditures imply that anticipated availability of research outcomes on the market will reduce the incentives to use only internally generated research outcomes. There are limits to relying on the market for research inputs into the development process. For example, internal research capability ... is needed to evaluate . . . [externally purchased research outcomes] ... and to synthesize them with other research outcomes, whether internal or external. But clearly some substitution takes place. If this analysis is meaningful, it suggests a division of labor according to firm size. Smaller firms will tend to specialize more in the research phase and in smaller development processes; larger firms will devote a much smaller proportion of their research and development budget to the research phase. They will specialize in larger developments and will buy a considerable fraction of the research basis for their subsequent development of innovations. (Arrow, 1983, pp. 26-7.)
But if this argument is correct, why did we not see a division of labor in innovation in the pharmaceutical industry for many years? Why do we see it only now? After all, patents have been a forceful instrument for protecting drug innovations at least since the end of World War II. In Arora and Gambardella (1993b and 1994b), we addressed these questions on a general basis. We argued that in many industries there has been a "technical" constraint upon what we called the "division of innovative labor." With limited understanding of the general and abstract laws of phenomena, highly experimental research produced information that was very context-specific. Research outcomes depended upon the local conditions in which experiments were conducted, and they could be used only for the specific purposes for which they were generated. Because it was difficult to extend technological information to other situations, the market for research outcomes was "thin", which reduced economic incentives to specialize in research in order to sell its findings. Moreover, the production of research outcomes depended primarily on learning, experience, and tacit capabilities. It was therefore difficult to articulate technological information in forms that could be usefully (and cheaply) transferred and used by other agents. Relatedly, it was difficult to describe it in forms that could be effectively patented or protected by intellectual property rights. Hence, research outcomes had to be used, for development and commercialization purposes, by the same agent that produced them, with the implied absence of a division of innovative labor. This also implied that large firms, which had established development, production, and commercialization assets, were then at an advantage, and they had greater incentives to invest in R&D. Smaller firms, with limited
78
Economic implications
downstream assets, could not utilize their research outcomes themselves, and they had fewer incentives to invest in research.20 With the growth of scientific knowledge and computational capabilities (especially over the last ten or twenty years), information for innovation, including information arising from experimental observations, can be increasingly related to different objects and contexts; hence, it can be used for different purposes, and it can be used by agents other than those who produced it. The market for research outcomes becomes "thicker". This enhances the economic advantages of specializing in research to sell technological information or other research outcomes (provided that intellectual property rights adapt). Also, as information for innovation can be related to more general frameworks, it can be articulated in forms that are more immediately intelligible to other parties. Not only can it be transferred at lower costs, but, because it can be articulated in more meaningful ways, it can also be protected more forcefully. A division of innovative labor between large firms and small innovative companies, along the lines suggested by Arrow, can then arise. Division of labor in drug research and innovation The pharmaceutical industry is a natural application of the framework discussed above. As argued earlier in this chapter, when drug discovery depended primarily upon extensive screening of molecules, and scientists had little knowledge about the human body and the action of drugs, the key assets for innovation were static and dynamic economies of scale. But scale and learning are indivisible assets. They can be created and utilized only in lumpy forms. Also, they can only be exchanged in lumpy forms (i.e., by selling the entire laboratory). Moreover, pharmaceutical companies had to have sales capabilities to generate the necessary cash flow to maintain large in-house research facilities, and to invest in extensive applied research screening. Thus, with established commercialization assets and financial resources, it was natural that they developed and marketed their own discoveries.21 With the recent advances in molecular biology and genetic engineering, drug innovations increasingly depend upon knowledge and information that are "generic" in nature, and that can be transferred at low cost among different agents. For instance, information about receptor structures can be comprehended by agents different from the ones who discovered them (provided that the "buyer" is knowledgeable in the field). Similarly, the properties of certain compounds, the action of drugs, or newly discovered activities of the human body can be expressed in fairly universal categories, and they can be transferred and used by other parties.22
Division of labor
79
More generally, unlike scale and experience, the knowledge-base for drug discovery has become more "divisible." With suitable contracts and intellectual property rights, relevant "fragments" of knowledge can be exchanged among specialized agents. A certain company or scientific institution can establish the structure of a family of receptors in the human body. Another party can clarify their biological activity. A third agent can determine the molecular action of a family of compounds that fit the receptor sites and counteract the undesired behaviour of cells. The entire set of information can be brought to a fourth party who controls resources to carry out long and costly clinical trials, and market the product. A division of labor in pharmaceutical innovation then becomes "technically" possible. (See also Arora and Gambardella, 1993c.) Large firms recognize that they can take advantage of knowledge, information, and product opportunities created by small innovative firms. Moreover, they can take advantage of the fact that, by resorting to an external market for ideas, they can "buy" only research outcomes that have shown some success. They save costs of unsuccessful projects, and they can shift at least part of the risk of initial research stages onto the small innovative firms. The latter, which have natural comparative advantages in producing ideas, realize that, with "divisibility" of science, they can invest in discovery, and sell their research outputs to larger firms. Patents and intellectual property rights will be very important to sustain the incentives of smaller firms to invest in upstream research.23 The rise of bio tech companies is also encouraged by the general applicability of their knowledge-base. Knowledge about the properties of certain drugs, and the way they act inside human receptors, can be employed for different purposes (e.g., different uses of the same family of drugs). Markets for knowledge are thicker, and small firms can supply their services to many clients.24 Moreover, a knowledge-based division of labor cannot be mediated by arm's-length market transactions (Arora and Gambardella, 1993b). The transfer of knowledge and other relevant information for drug research and innovation requires longer-term contracts and tighter interactions, like joint ventures, research contracts, and similar arrangements. The transfer of knowledge requires mutual learning and the exchange of complementary tacit information (Arora, 1991). It also requires mutual trust. Furthermore, fruitful relationships often depend on the ability of joining resources and knowledge-bases to carry out specific projects, rather than on the mere delivery of an idea, or compound, from a biotechfirm,or a university, to a large firm. Finally, tighter relationships between large pharmaceutical manufacturers and smaller biotech companies may be necessary to relieve the latter of part of the risk of initial research stages. Through collaborative
80
Economic implications
agreements, small capital investments in innovative firms, or other relationships, larger companies would provide financial and organizational support. Small innovative companies might be unable to bear the entire risk of their research efforts, and thus they might not engage in research in the first place. Although large companies would now bear some of the risk of starting research (with respect to pure arm's-length market transactions), they would nonetheless bear lower risks than with full internalization of these projects, and they could take advantage of the greater abilities of the smaller companies in earlier research stages. The changing attitude of large pharmaceutical companies towards a division of labor in innovation
The formation of many small-medium research-intensive biotech companies in the 1980s might suggest that present network relationships in the pharmaceutical industry are driven by an increased supply of specialized research skills. In fact, established pharmaceutical companies display increasing demand for such collaborations. This is important, as it is sometimes argued that the present wave of network arrangements in the drug sector is a transitory phenomenon, prompted by a change in the technological paradigm. But as large firms realize the advantages of a division of labor in innovation, the new trend may become a more persistent feature of this industry. As the previous discussion suggested, if a market for technological ideas arises, the largefirmsthemselves willfindit advantageous to use this market instead of relying entirely on their internal research.25 Large firms show strategic and organizational inertia (Henderson and Clark, 1990; Pavitt, 1992; Henderson, 1993), and pharmaceutical corporations are no exception (Sharp, 1991b). Sapienza (1989a) argued that established drug companies exhibit "bureaucratic stolidity," which brings about managerial risk aversion rather than innovation. But some pharmaceutical multinationals are modifying their innovation strategies, and they are recognizing the benefits of a division of innovative labor. One could easily list the vast number and the heterogeneous forms of agreements of major pharmaceutical corporations worldwide. Tables 3.23.5 reported the external linkages of four representative companies. As suggested earlier, many large US and European (and even Japanese) drug corporations have established linkages of similar breadth and intricacy. Moreover, chapter 7 will present careful statistical analysis of the propensity of large companies to develop external relationships in this industry. Here we look instead at the recent position of some large pharmaceutical firms. Although they provide somewhat scattered evidence, these examples elucidate the important changes in the views of drug corporations about
Division of labor
81
innovation strategies and knowledge-sharing with partners. (See also FT, 1989a.) Some pharmaceutical multinationals have recognized their "bureaucratic stolidity," and have undertaken profound reorganizations to encourage creativity and risky innovations. Alex Kramer, the Chairman of CibaGeigy, has declared that his company "should no longer navigate as a supertanker, but as a flotilla with independent captains on every ship" (Scrip, 1991b, p. 15).26 More importantly, large companies are changing their attitude towards external partnering. William Parfet, President of Upjohn, declared that the only way pharmaceutical firms can afford the increased R&D costs is to form strategic alliances with academia, technology start-up firms, and even with competitors with complementary R&D strengths and strategies (MKTLTR, 1992). Similarly, as we shall see in more detail in the next chapter, Eli Lilly recognized its narrow vision about full internalization of resources for innovation. It is now hiring high-profile scientists from research institutions, it is trying to combat its "not invented here" syndrome, and it is encouraging systematic partnerships with research-intensive companies (BW, 1992b). Finally, Sandoz set up a 150 million dollar program, called Innovascan, which sponsors promising new lines of research suggested by internal divisions or in collaboration with external parties. The program plans to support the various projects, including collaborations, from research to the marketplace (Scrip, 1991a). An official declaration of Max Link, the CEO of Sandoz Pharma Ltd., epitomizes the changing attitude of pharmaceutical multinationals: "We believe that the stronger a multinational is in a particular field, the higher the probability that a co-operation will yield positive results" (Scrip, 1991a, p. 7). The new philosophy of pharmaceutical corporations is also exemplified by some recent remarks of Jurgen Drews, Director of Corporate Research at Hoffmann La Roche (MKTLTR, 1991). He emphasized that even large multinationals have to abandon the idea of covering an exhaustive range of research areas. The key for success is to specialize in selectedfields,and to rely on partners for complementary research and specialized technical expertise. He also argued that individual competitiveness will be replaced by competition among networks of firms based on trust, credibility, and openness of research. More recently, Drews discussed the transformation of the market structure of the international drug sector (Drews, 1992). He predicted that the industry will polarize around two fundamental agents: giant multinational corporations with considerable abilities to manage large and complex organizations, and small-medium firms with sophisticated scientific expertise in selected areas, which will act as "suppliers of ideas, product opportunities and new technologies" (Drews, 1992, p. 138).
In-house scientific research and innovation: case studies of large US pharmaceutical companies
Introduction
As argued in the previous chapters, there are important differences in the ability of firms to utilize scientific knowledge. As scientific knowledge influences the most creative stages of drug development, this can translate into notable differences in the innovation and market performance of drug companies. Moreover, these differential abilities can be magnified by the growing scientific intensity of pharmaceutical research: The scientific ferment ... is separating the drug companies in terms of capability much more than they used to be. Some are going into much more imaginative processes to look for new drugs, others are still conducting things the way they used to. The sky is indeed bluer, but only for some. (Dr. Peter Goldman, Professor of Pharmacology, Harvard University, quoted in BW, 1979, p. 137.) Pharmaceuticals are changing... from a business in which everyone succeeds to one in which success is more selective ... [T]here will be a greater divergence in individual performances. (Eugene L. Step of Eli Lilly, quoted in BW, 1979, p. 145.) This chapter examines whether, in spite of the public nature of science, large US pharmaceutical firms differ in their ability to exploit scientific knowledge, and whether this implies differences in their innovation and market performance. The analysis is based upon case studies of some major US drug manufacturers. The case studies show that the most successful US pharmaceutical firms during the 1980s were those that invested systematically in scientific research. This enabled them to generate scientific knowledge internally, and to exploit external knowledge more successfully. The case studies also show that better capabilities of exploiting science, and hence better innovation and market performance, are associated with firms that organized, at least in part, their internal research in ways that encouraged a scientifically creative atmosphere through open research and linkages with the scientific community. 82
Merck
83
The case studies also highlight the self-reinforcing properties of the science-innovation spiral in the pharmaceutical industry. Firms with better scientific capabilities have been able to produce a systematic flow of new drugs. They have accumulated internal funds that were reinvested in science, thereby strengthening their competitive position. Firms with limited in-house scientific expertise have faced serious difficulties in becoming major innovators in this industry. The case studies cover the following firms: Merck, Eli Lilly, BristolMyers, Squibb, Smithkline, Syntex, American Home Products, and Rorer. It can be seen from table 4.1 that they include the first four firms in terms of 1986 US pharmaceutical sales, and five out of the first six. Overall, the eight firms above cover 36 percent of the 1986 US pharmaceutical sales of the top fifty companies in this market (MAN, 1987). The case studies provide a representative sample of the different strategies of the largest US pharmaceutical companies. Merck and Eli Lilly are highly research-intensive firms with strong in-house scientific capabilities. Bristol-Myers has strong marketing assets and a strong competitive position in non R&D-intensive products; it invested heavily in research during the 1980s to enter the market for patented drugs. Squibb, Smithkline and Syntex have good in-house research; their marketing position, however, relied upon the sales of one major product ("one-drug companies"). American Home Products is another marketing giant with a strong position in non R&D-intensive products. Rorer is a medium-sized firm with modest in-house research; during the 1980s, it attempted to expand its research operations to break into the market for R&D drugs. By and large, these stories cover all the interesting cases for our purposes. One could add cases of other major US pharmaceutical companies (e.g., Pfizer, Upjohn, Warner-Lambert). However, they would only reiterate the points developed for the firms that have been investigated. The following sections present each case study in turn. There is also a concluding section presenting the lessons from the case studies. Merck Merck was the most successful pharmaceutical firm during the 1980s. It introduced a host of new drugs. It also has a number of compounds in late clinical trials, which suggests that its competitive position is likely to be reinforced in the near future (see table 4.2; see also CMR, 1987b, and UST, 1987). Merck's high performance rested upon superior in-house research expertise. This, in turn, is associated with an organization of internal research that stimulates scientific creativity, and that resembles that of
Table 4.1. 1986 pharmaceutical sales and 1983-88 average R&D/sales ratio of the eight firms in the case studies Company
Rank of company in 1986 US pharm. sales
1986 US pharm. sales (million $)
1986 world pharm. sales (million $)
Average R&D/sales ratio (percent)
American Home Prods. Merck Smithkline-Beckman Bristol-Myers Eli Lilly Squibb Syntex Rorer
1 2 3 4 6 11 19 23
2,053.0 1,749.0 1,626.3 1,592.5 1,493.8 1,016.9 537.5 471.0
2,933.3 3,441.0 2,502.2 3,598.0 2,120.0 1,529.6 803.2 845.0
4.4 11.3 9.8 5.9 11.8 10.8 14.7 7.1
Source: 1986 sales data and company ranking from MAN (1987); average R&D/sales ratio from R&D and sales data in Moody's Industrial Manual (various years).
Table 4.2. Merck: major new drugs in the market or in the pipeline, late 1980s Drug
Category
Year of market approval/status
Primaxin Pepcid Vasotec
Antibiotic H2 antagonist ACE-inhibitor (
1985 1985 1986
Mevacor
Anti-cholesterol
1987
Prinvil Losec
ACE-inhibitor Anti-ulcer
1988 1990(?)
Proscar
Benign prostate tumor
1992
MK 538/Prodiac
Aldose reductase inhibitor
In late clinical trials
Source: BS&C (1989d), and other trade magazine sources.
Notes
Second ACE-inhibitor drug in the market after Squibb's Capoten First anti-cholesterol drug; spectacular sales growth in first two years of marketing Second-generation Vasotec; marketed with ICI Expects approval also for therapies other than ulcer (Zollinger-Ellison syndrome and gastroesophageal reflux disease) Expected to become Merck's third one-billiondollar drug by 1994
86
Case studies of US pharmaceutical companies
academic departments.1 Merck's laboratories are very informal, and scientists can easily be in touch with one another, which is explicitly meant to attract top-ranking scientists from university, and make them feel comfortable in an academic-like environment. Moreover, Merck's scientists enjoy unusual freedom. They often take autonomous decisions about research lines to pursue, and on many occasions they investigate topics of personal interest (BW, 1987). Research at Merck is divided into therapeutic areas, which are organized into projects. Projects typically correspond to compounds that have shown some promise in early experimental stages. Each project, including apparently successful ones, has no budget granted by authority. This is an interesting feature of Merck's research organization. If the head scientist of a particular project believes that her work needs additional resources, she has to convince researchers in otherfieldsor projects to commit part of their budgets and time to her program. Resources are thus allocated according to the scientists' evaluation of different research lines (BW, 1987; UST, 1987). This model bears interesting similarities with peer-group evaluation in the scientific community. Moreover, within the scientific community, when early breakthroughs open new research opportunities, many scientists rush into the new areas. They are lured by the pecuniary and non-pecuniary rewards that may derive from research in relatively unexplored directions, which exhibit higher potential for discovery. Thus, the allocation of scientists' time and resources is chosen by the scientists themselves, who are the best "judges" of the scientific potential of different research programs. Similarly, company scientists at Merck enjoy some discretion (albeit probably not full) to move across projects, and therefore to contribute to projects that they regard to be of some value. Their incentives to move into potentially successful projects are, of course, the pecuniary and nonpecuniary rewards that may arise from being part of a fruitful research endeavor. Merck's research organization also fosters systematic relations with the scientific community. Merck's library has a reputation comparable to the best academic centers. The company regularly organizes internal seminars of the world's top academic scientists.2 In addition, Merck's scientists are encouraged to establish autonomous linkages with academic institutions, even for projects not officially approved by top managers.3 The story of Mevacor, the anti-cholesterol drug marketed in 1987, illustrates Merck's scientific skills in drug discovery, and the effectiveness of a research organization that draws in many ways from the model of academia. As discussed in chapter 2, Mevacor is a major example of discovery by design. Merck scientistsfirststudied how cholesterol is formed within the human body. They then searched for a "weak" link in the peptide
Merck
87
chain that produces cholesterol. They found an enzyme that, if inhibited, could block the production of cholesterol. Finally, they searched for a compound with suitable characteristics that could inhibit the action of the enzyme. Mevacor's story also highlights Merck's ability to build upon publicly available science. Although mevalonic acid, a chemical link in the cholesterol chain, was first isolated by Merck's scientists in 1956, research on anticholesterol drugs was spurred by later findings. Between 1972 and 1974, Michael S. Brown and Joseph L. Goldstein of the University of Texas identified the key steps in the production of cholesterol, work for which they were awarded the Nobel Prize in 1985. Brown and Goldstein's findings motivated Merck's scientists to launch research on cell culture assays for cholesterol inhibitors as early as 1975. In 1978, Merck isolated lovastatin, the Mevacor compound, from a micro-organism of the soil. Mevacor's NDA was approved for marketing in August 1987. The product reached $260 million sales in 1988, the first full year of marketing, and it reached $1 billion sales in 1991 (BW, 1987; FDA, 1988; BS&C, 1989d; FT, 1992a). As soon as their discovery had been made, Brown and Goldstein's achievements became publicly available. Yet, Merck was the only company that effectively exploited their findings. Other firms had interests in the anticholesterol business. Bristol-Myers, for instance, already had an anticholesterol product, Questar. As we shall see later, in the early 1980s Bristol-Myers did not have significant in-house scientific expertise. Unlike Merck, it was unable to use the new (publicly available) discovery to improve Questar, and it has not yet come up with a major new anticholesterol product.4 Mevacor's success also depended upon strict ties between research and marketing. In 1979, soon after the synthesis of lovastatin, Merck's scientists gathered with marketing managers to assess the commercial potential of an anti-cholesterol product. They wanted to know the market opportunities of a new anti-cholesterol drug before committing large fractions of the company's research time and expenditure to the project. Marketing managers suggested that the market for anti-cholesterol drugs was wide open. Moreover, early contacts with marketing managers helped in perfecting some characteristics of the drug. The marketing staff informed the scientists that Bristol-Myers' Questar, the major product in the market, was unpleasant to swallow. Merck's scientists then directed part of their research to make the new drug more palatable. Although not a key feature of the product, this proved to be a non-trivial factor in promoting its market diffusion (BW, 1987). Apart from Mevacor, in 1985 the FDA approved Vasotec, the second ACE-inhibitor anti-hypertensive drug after Squibb's Capoten. By 1988,
88
Case studies of US pharmaceutical companies
Vasotec had already captured 46.6 percent of the hypertensive market against Capo ten's 49.3 percent. Merck has also launched Pepcid, an antiulcer drug which is competing against the "old" anti-ulcer compounds Tagamet of Smithkline and Zantac of Glaxo, as well as with the newcomer Axid of Eli Lilly. In 1992 it launched Proscar against benign enlarged prostate in men. Proscar rationally inhibits an enzyme that is responsible for enlargement of the prostate, a condition that affects millions of men. Proscar has very good commercial expectations. It is expected to become Merck's third one-billion-dollar drug after Mevacor and Vasotec.5 Merck has also capitalized on its previous research successes. At present, it is developing Zocor, the second-generation Mevacor, and it is launching (jointly with ICI Chemicals of the UK) Prinvil, the second-generation Vasotec. It is also launching Losec (with the Swedish company Astra), a new anti-ulcer drug. Losec, which exploits previously accumulated knowledge of anti-ulcer drugs (which led to the introduction of Pepcid), is also believed to be effective in treating the Zollinger-Ellison syndrome (a hypersecretory disease) and a gastro-esophageal reflux disease. Although Proscar is pioneering an entirely new market for Merck, its development relied on Merck's expertise in designing enzyme inhibitor drugs (BS&C, 1989d; NYT, 1992a and 1992b). Finally, Merck's high-quality research and its experience in drug development have enhanced its reputation with the FDA. Mevacor was approved by the FDA in the record time often months (BW, 1987; FDA, 1988). Similarly, Prinvil was approved in twenty months, about 10 months less than the average waiting time (BW, 1988d). Development of many new drugs has helped Merck in acquiring good expertise about how the FDA wants the clinical trials to be conducted, and how to properly file an NDA. Moreover, Merck has a staff of 120 people to organize information from laboratory and clinical trials and to prepare NDA reports {BW, 1987). The company pays a great deal of attention to the structure, format, and composition of NDA reports, which is of great help in speeding up the revision process. Eli Lilly Eli Lilly is another research-intensive firm, with a long-standing tradition of drug research and innovative performance, especially in antibiotics. It was one of the first corporations worldwide to undertake biotechnology research, as early as the mid-1970s (OTA, 1984; Pisano, Shan, and Teece, 1988). This translated into a successful early entry into biotechnology. By 1985 Eli Lilly had already commercialized two biotechnology-based human therapeutics, human insulin (the first biotech drug to reach the
Eli Lilly
89
Table 4.3. Eli Lilly's drugs with 1989 sales of at least $100 million Drug
Category
1989 sales (million $)
Ceclor Prozac Humulin Keflex-Keftab Animal insulin Dobutrex Tylan Vancocin Darvon Axid Nebcin
Oral antibiotic Anti-depressant Human insulin Oral antibiotics
715 350 300 190 165 150 145 130 125 100 100
Heart failure drug Injectable antibiotic Anti-ulcer Injectable antibiotic
Source: NYT (\990g). market, which was developed and commercialized jointly with Genentech) and the human growth hormone. 6 Eli Lilly also ranked second among all institutions, and first among companies (including large firms and biotech companies), in terms of US patents in genetic engineering granted by December 1987 (twenty-seven patents) (OTAF, 1987).7 Eli Lilly is investing in computer-based molecular modeling, and by the end of the 1980s it had already planned to install a supercomputer to design complex molecular structures (CW, 1989e). In 1988, it signed an agreement with Agouron Pharmaceuticals, a small biotechnology firm specializing in three-dimensional computerized design of drugs. The agreement involves joint development of new compounds using three-dimensional computer techniques. Lilly is "learning" the new technique from the specialized concern, and it has first manufacturing and marketing rights on new products in exchange for multi-year funding of Agouron research (Bioscan, 1988; CW, 1988a). Table 4.3 summarizes Eli Lilly's market performance in the 1980s. In 1989 it had eleven drugs with sales greater than $100 million. Among them, Prozac and Axid, an anti-ulcer drug approved in 1988, are expected to be its major products in the 1990s. As discussed in chapter 2, Prozac, an anti-depressant approved in December 1987, is an important example of discovery by design. Its development hinged upon vast research efforts, mostly in the public domain, especially since the mid-1970s. Such research efforts were prompted by a new basic understanding of protein receptors, and particularly of the chemistry and action of serotonin. In the 1980s, many
90
Case studies of US pharmaceutical companies
companies invested in serotonin drugs. However, Prozac was the first major breakthrough in this field. Eli Lilly's scientists effectively exploited the public information that depression is triggered by quick absorption of serotonin by nerve cells. In-house research expertise was critical for the careful design and development of a drug that attenuated depression by slowing the absorption of serotonin. It was also critical for monitoring and utilizing public knowledge. Lilly's scientists did not develop a new map of the structure of brain receptors, nor did they produce original knowledge about new receptor sites or other characteristics of serotonin receptors. But they utilized this knowledge effectively to make an important discovery.8 Moreover, serotonin appears to have numerous potential effects, and Eli Lilly is trying to capitalize on its early entry in this area. Apart from careful monitoring of public research (which is revealing new effects of serotonin, and new serotonin receptors), Lilly's scientists have noted that Prozac spurs weight loss, and they are studying its action against obesity. Paradoxically, they have also observed that, with suitable modifications, Prozac could treat the opposite problem, anorexia (BW, 1988a). Lilly has also signed a $12 million contract with Synaptic Pharmaceutical, a small company specializing in serotonin research. Synaptic has cloned a number of receptors, and it is helping the company in developing serotonin drugs for anxiety, depression, etc. The agreement has already produced four candidate serotonin drugs, which are now in animal tests (BW, 1992a).9 In spite of its solid knowledge-bases and research successes, Eli Lilly's competitive prospects for the 1990s have shown some signs of weakness. Prozac is facing increasing competition, and in 1992 its sales did not grow as expected (BW, 1992b). Also, a number of firms with sound in-house scientific bases are entering the field of serotonin drugs (among others, Merck, Smithkline, Glaxo, Janssen Pharmaceutica; see BW, 1992a; see also chapter 2). Moreover, in December 1992 Eli Lilly's patent on its antibiotic Ceclor, one of its key products, expired. Generic manufacturers have already applied to the FDA for competitive products. More importantly, in 1992 Eli Lilly's profits declined considerably (BW, 1992b). Eli Lilly is reacting through a profound reworking of its strategy (BW, 1992b). Lilly's action appears to match rather closely the discussion in chapter 3. Vaughn Bryson, the new CEO, is making significant steps to transform Eli Lilly's rigid, bureaucratic organization into a far more decentralized decision-making structure.10 Bryson is also hiring scientists from outside the company. Eli Lilly has now hired high-profile researchers from institutions such as the National Institute of Health and the Center for Disease Control. Executives hope to gather a significant group of talented scientists to revamp Lilly's research. More generally, they are taking serious steps to make both managers and researchers more receptive to outside ideas.
91
Eli Lilly
Table 4.4. Eli Lilly's research alliances in the early 1990s Lilly's investment (million $)
Company
Project
Centocor
Developing a drug to combat massive bacterial infections
Repligen
Hoping to produce drugs to treat trauma shocks and respiratory distress
14.0
Synaptic Pharmaceut.
Research on serotonin-based drugs
12.0
Oclassen Pharmaceut.
Developing an oral drug to treat hepatitis B
7.5
Shaman Pharmaceut.
Working on drugs to treat respiratory virus and herpes
4.0
Sibia
Developing drugs to treat nervous-system diseases such as Alzheimer's
4.0
Bioject
Developing a system to inject medication through the skin without a needle
4.0
100.0
Source: BW (1992a and 1992b).
Lilly also recognized that its previous successes produced an excessive "not-invented-here" syndrome. The company has changed dramatically its attitude towards external partnering. Since 1991 it has signed agreements with a number of small companies, and acquired minority stakes and marketing rights for products in development (table 4.4). Such alliances are an obvious recognition of the advantages of the division of innovative labor with smaller research firms: "Bryson is . . . expanding an ambitious plan to forge strategic alliances with young research companies that he hopes will pump new products in Lilly's pipeline" (BW, 1992b, p. 72). Of particular importance is the agreement signed in 1991 with Centocor to revamp development of Centocor's compound Centoxin, which treats bacterial infection. Although grounded on a solid research basis, Centoxin's NDA was rejected by the FDA, which cited poor test results. This was an important setback for Centocor, which hoped that Centoxin's commercialization would have helped it in becoming a fully integrated pharmaceutical manufacturer. Eli Lilly offered its greater expertise in drug development to properly meet FDA requirements. Lilly also offered financial
92
Case studies of US pharmaceutical companies
resources to enable Centocor to perform additional research that would prove Centoxin's worthiness to the regulator. Finally, Lilly will supply its extended commercialization network for Centoxin sales. In short, while Centocor had considerable research capabilities, Lilly offered better downstream resources. More importantly, both companies have recognized each other's specialization, and they are taking advantage of division of labor instead of seeking forward or backward integration (BW, 1992b). Bristol-Myers/Squibb In 1989, Bristol-Myers and Squibb merged to form a new giant corporation. During the 1980s, the two firms had fairly diverse, but "partially" successful stories. The merger suggests that neither's strategy was sufficiently effective to cope with the rising competitive pressures of the US pharmaceutical market in the 1990s. Before the 1980s, Bristol-Myers was specializing in health care and consumer products. It had strong marketing capabilities, but a modest research base. In the 1980s, it took major steps to become a researchoriented group. It started an extensive program to reorganize internal research. In 1986, it concentrated its research laboratories, previously scattered in three separate US locations, in one new research center in Connecticut. The location was chosen because of its proximity to Yale with which it had signed a broad research agreement in 1982 (see below) and other universities of New England. This was explicitly meant to facilitate informal contacts between academic and industry scientists (CMR, 1986; WSJ, 1988a). During the 1980s, Bristol-Myers also made important in-house research investments in genetic engineering and molecular biology (WSJ, 1988a). Moreover, Bristol-Myers entered into research collaborations with various academic institutions.11 Particularly, in 1982, it signed a comprehensive agreement with Yale University on anti-cancer research. The agreement, which was renewed in 1986, covered a broad number of scientific disciplines, and established that Bristol-Myers financed research at Yale in exchange for a first option on licenses. (See table 3.1; see also JC, 1988.) The company also acquired research capabilities in biotechnology via acquisitions. In 1985 and 1986, it acquired Genetic Systems and Oncogen, two important medium-sized firms specializing in biotechnology research (Bioscan, 1988; WSJ, 1988a). The new strategy produced some results. Bristol-Myers now performs frontier research in a number of fields, including sophisticated basic research in the area of nerve and brain cell chemistry, and in biotechnology (WSJ, 1988a). However, in spite of this successful shift towards research, its
Bristol-Myers/Squibb
93
market performance did not improve considerably during the 1980s. The company was unable to market a fundamentally new drug, nor were radically new compounds in the pipeline. The sales of its major product, Buspar, a tranquillizer, were not as successful as expected. Moreover, Bristol-Myers was unable to capitalize on its experience in anti-cholesterol drugs. Bristol-Myers pioneered the field with Questar. Yet, Merck's Mevacor has now taken the lead. Bristol-Myers' story suggests that rapid catch-up strategies are not sufficient to become a leading innovator in this industry. Apart from enormous research costs, clinical trials may well take a decade before new drugs can be marketed. The ten-year time span since Bristol-Myers reorganized its internal research is probably only a fraction of the time actually needed for this strategy to pay off. Moreover, scientific expertise has to be paired with experience in drug development. We saw that Merck's scientific reputation and its expertise in drug development have favored a climate of mutual trust with the FDA, which has often streamlined Merck's drug approvals. In contrast, Buspar's NDA was blocked by the FDA for more than one year. The FDA claimed that the company had not properly conducted two important clinical studies during pre-marketing trials on patients (WSJ, 1988a). Squibb's story during the 1980s is different from Bristol-Myers'. The company had a good research-base. Squibb's success during that decade rested upon the discovery, development and marketing of the first angiotensin-converting-enzyme (ACE) inhibitor anti-hypertensive drug, Capoten. The origins of Capoten were in the 1960s, and involved two different lines of research (WSJ, 1987b). The first line of research was concerned with the genesis of hypertension. Researchers working on hypertension realized that renin, a chemical released by the kidneys, causes blood to produce another chemical, angiotensin I, which in turn produces angiotensin II. Angiotensin II is the ultimate regulator of blood pressure. Overproduction of angiotensin II is a major cause of hypertension as it sharply increases blood pressure. Scientists then realized that they had to find a drug that blocked the action of the enzyme responsible for releasing angiotensin II. The second line of research was related to the cause of death from the venom of the Brazilian viper. Researchers found that the victim dies because the venom contains an extract that lowers blood pressure by inhibiting the production of angiotensin II. Thus, understanding the chemical structure of the venom offered the opportunity of re-producing a compound for treating hypertension. John Vane, a British pharmacologist, brought these findings to Squibb. The company was working on heart medicines. Building upon previous
94
Case studies of US pharmaceutical companies
knowledge about the causes of hypertension and the action of the viper's venom, Squibb's scientists constructed a molecule that mimicked the function of the viper's compound. Captopril, the chemical compound of Capoten, effectively blocked the enzymes responsible for the conversion of angiotensin. Capoten was first approved for marketing in 1981 to treat severe hypertension, and then in 1985 to treat milder forms. Once again, the discovery was spurred by publicly available scientific knowledge. Knowledge about angiotensin conversion, the properties of the chemicals that influence blood pressure, as well as the properties of the viper's venom were public information. Squibb had adequate in-house research capabilities to take advantage of the information. First, it was performing scientific research in related domains. This gave it a lead in monitoring external research and in recognizing the potential of the new advances. Second, Squibb's skills in drug design enabled it to use the external knowledge to generate a compound with unique therapeutic properties. Squibb tried to capitalize on its expertise gained with Capoten's research. It started rational investigation of new uses of Capoten, which led to its approval as a treatment for congestive heart failure, a problem afflicting some 3 million Americans and therefore likely to give additional impetus to the drug's sales. In addition, Squibb has begun studying Capoten as a means of preventing heart attack, which could considerably boost its market performance. In the late 1980s, Squibb also had other drugs in the pipeline. Apart from three new ACE inhibitors and an anti-inflammatory drug, it had high expectations about a new anti-cholesterol drug, Pravachol (BW, 1988d). Despite these actual and potential opportunities, Squibb's market position in the 1980s was not completely secure. Squibb was a "one-drug company." Capoten accounted for about 40 percent of its sales {BW, 1988d). The risks for one-drug companies are apparent. As competitors erode the market shares of their main product, a substantial fraction of their profits can be affected. In effect, Capoten faced increasing competition from Merck's Vasotec and Prinvil. This posed no immediate threat to the company. The market for anti-hypertensive drugs was large enough to sustain three products. However, Squibb clearly needed to escape its onedrug-company status. This was necessary to gain greaterfinancialstability, and to plan long-run research. The 1989 merger integrated two important complementary assets. Apart from the cashflowgenerated by Capoten, Squibb brought major research capabilities. Bristol-Myers brought marketing skills and an extended marketing network. Bristol-Myers' marketing assets will be particularly valuable to support Capoten's sales. They will also be critical in promoting
Smithkline
95
the commercialization of Squibb's new drug Pravachol.12 Moreover, Bristol-Myers' financial resources, which stem primarily from its health care and consumer product business, will help Squibb escape the problems typically associated with one-drug companies. They will supply the necessary stability for further research, thereby diminishing the risks of relying only upon the financial flows generated by Capoten. Particularly, BristolMyers' financial resources have enabled the new company to invest in research on brain-related disorders, a field that Squibb was planning to investigate but lacked adequate funds. Bristol-Myers also supplied a good research-base in anti-cancer, where Squibb had only just started new research programs (WSJ, 1989d). Smithkline Smithkline's story during the 1980s is largely associated with Tagamet, the first H2-antagonist anti-ulcer drug, which it introduced to the market in 1977. The company enjoyed a virtual monopoly in the anti-ulcer market between 1978 and 1983. In 1983, Glaxo, a British company with a modest presence in the USA at the time, introduced a competing product, Zantac. Zantac had a few advantages over Tagamet, such as a twice-daily dose against the four times of Tagamet, and less severe side-effects. Moreover, Glaxo undertook an aggressive marketing strategy. In a few years Zantac overtook Tagamet to become the leading anti-ulcer drug (ECON, 1986; BW, 1988c). Smithkline reacted by attempting to improve Tagamet. It successfully reduced its dosage to twice per day. Moreover, Smithkline sought to develop various "sons" of Tagamet to diminish its side-effects. These attempts proved to be unsuccessful, and the company has not been able to produce a significantly improved version of the drug (BW, 1988c; WSJ, 1988b and 1989a). In 1988, Tagamet's sales fell by 16 percent. Smithkline was another onedrug company. Tagamet accounted for about 25 percent of its sales and 40 percent of its profits. The drop in Tagamet's sales had a major impact on Smithkline's financial performance. Moreover, in 1987, its patent of Dyazide, a blood pressure drug, expired. Dyazide sales dropped by 51 percent because of competition from manufacturers of generic products. In 1988, Smithkline's net income fell to $229.2 million from $570.1 million in 1987 (BW, 1988b; Moody's Industrial Manual, 1989; WSJ, 1989a). In addition, Smithkline's future prospects were uncertain. In the late 1980s, it had twenty-eight drugs under development, compared with a handful in the early years of the decade. Moreover, in the 1980s it marketed a genetically engineered vaccine against hepatitis B. Yet, its most important
96
Case studies of US pharmaceutical companies 10.5-
CO
Q
6.5
80
85
Fig. 4.1. Smithkline's R&D/sales ratio, 1959-88. (Source: R&D and sales data for 1959-85 from NBER-Compustat files [Hall et al., 1988]; for 1986-88 from Moody's Industrial Manual [various years].) new drugs had not performed as expected. The sales of its new antibiotic, Monocid, were modest. Its new anti-arthritic drug, Ridaura, showed nontrivial side-effects in clinical tests, which dampened its potential for sale {Bioscan, 1988; BW, 1988b; WSJ, 1989a). Some industry analysts suggested that Smithkline's low performance depended on its inability to establish major in-house research capabilities during the 1980s (BW, 1988b and 1988c; WSJ, 1989a). Figure 4.1 reports the ratio of Smithkline's R&D expenditures to sales between 1959 and 1988. The ratio peaked during the late 1960s and the early 1970s. These were the years of Tagamet's research, and a period of strong scientific and intellectual ferment at the Welwyn (UK) research laboratory, where Tagamet was discovered (Sapienza, 1987). The ratio declined immediately after 1974. Tagamet was under regulatory revision for marketing approval, and the bulk of research on the product had ended.13 The ratio, however, started rising again in 1979, when Tagamet's sales increased sharply, and it rose all the way to 1988. The company appears to have reinvested its profits in research, which suggests that Smithkline's problems did not really arise from its inaction in translating Tagamet's surplus into new research. In fact, Smithkline's poor research performance can be ascribed to the type of research investments it carried out in the 1980s. Particularly,
97
Smithkline 800 n EH Lilly + Merck O Smithkline
700 600 CO
500 &
2
400
_Q
300 200 100
65
70
Fig. 4.2. Scientific papers published by company scientists: Smithkline, Eli Lilly, and Merck, 1964—88 (number of papers from a constant 1981 journal set of Science Citation Index). (Source: CHI Research/Computer Horizons and Science Citation Index [various years]. See appendix to chapter 5 for further detail on sources and construction of data.) Smithkline did not reinvest any substantial portion of its profits in upstream research. Figure 4.2 shows the 1964-88 trend in the number of scientific papers published by Smithkline's scientists, and compares it with Merck's and Eli Lilly's. Although publications are only an approximate measure of in-house scientific capabilities, they can be taken as evidence of the extent to which company scientists are plugged into the scientific network. Between 1975 and 1985 Smithkline had systematically fewer papers than Merck and Eli Lilly. Merck showed a significant upward trend. Eli Lilly had about 100-120 papers per year (in a constant 1981 journal set). Smithkline scientists published about 50 papers per year up to 1981, and only in 1985-88 do we observe a major upswing from below 100 to about 150-200 papers. Apart from papers, the two scientists responsible for Tagamet's success, James Black and William Duncan, left the company immediately after Tagamet's research ended. Black left in 1973, Duncan in 1979. Black and Duncan were quite valuable assets. Not only did they lead the Tagamet project, but they also helped in creating an environment especially congenial to scientific research. When Black and Duncan resigned, the company did not strive to hire other leading scientists that could replace not only
98
Case studies of US pharmaceutical companies
their research skills, but also their capabilities in furthering a stimulating atmosphere. Tagamet's profits were used, for instance, to acquire Beckman Instruments in 1982. Beckman had a good research basis in instrumentation. The acquisition, however, was to a large extent an investment alternative to scientific research. In sum, Smithkline failed to use the proceeds and the market position acquired with Tagamet to establish as early as the mid-1970s a strong inhouse scientific research basis, or at least to continue the research tradition initiated by Black and Duncan. As suggested by the rise in scientific papers between 1985 and 1988, Smithkline did attempt to boost its in-house scientific capabilities after Tagamet had been on the market for some years. Moreover, the 1988 drop in profits prompted executives to undertake a major reorganization of research. They concentrated various research divisions in one major facility. They also sold part of the stakes in Beckman Instruments to raise funds for research (BW, 1988b; WSJ, 1989a) - which also suggests that this was not the best allocation of Tagamet's surplus back in 1982. These initiatives, however, appear to have been spurred by external events. The rise in scientific papers during 1985-88 corresponds to the first threats from Glaxo's Zantac. The reorganization of research followed evident needs for restructuring after the drop in performance. In a sector like drugs, where scientific research pays off after a decade or so, these moves were unlikely to yield immediate commercial results. Tagamet had a virtual monopoly in the anti-ulcer market for five years. Smithkline could have established strong in-house scientific research capabilities before Glaxo entered the market. It did not find such stimuli internally, and waited until competitive pressures rendered it necessary to accomplish a profound reorganization of research. Paradoxically, Smithkline's story corroborates the view that market success in the research-intensive ethical drug business is a self-reinforcing process. Previous successes bring about resources to afford long-run investments in scientific research, which sets the ground for future successes. Smithkline failed in one link of this chain. It did not exploit past successes to build a solid in-house scientific research basis, and this undermined its capabilities for enjoying cumulative advantages. In 1989, Smithkline merged with the British pharmaceutical company Beecham. The new company is another giant corporation in the drug business, with significant research and marketing capabilities. Among other things, Beecham has supplied the necessary financial stability after Smithkline's decline in performance following its difficulties in building on its early success with Tagamet.
Syntex
99
Syntex
Syntex is another one-drug company. It produces Naprosyn, a nonsteroidal anti-inflammatory drug based on the compound naproxen. Naprosyn, which was first marketed in 1976, is one of the world's topselling drugs. At the end of the 1980s, it had yearly sales of around $700-800 million. 14 As with all one-drug companies, Syntex's major concern is that its overall performance depends on one product. Naprosyn is experiencing a serious competitive challenge from Ciba-Geigy's Voltaren, which was introduced in the USA in late 1988, and which is gaining market share against Naprosyn. In 1989, Upjohn introduced another non-steroidal anti-inflammatory drug, Ansaid, which poses additional competitive threats {BS&C, 1988 and 1989f). Syntex also faces potential competitive threats from generic products that may flood the market after the expiry of Naprosyn's patent in 1993. Syntex is reacting to Naprosyn's patent expiration. It is developing new production technologies to manufacture Naprosyn at competitive prices after 1993. It hopes to meet competition from generic products by cutting Naprosyn's price. It now produces Naprosyn infiveplants which use batch processes. Syntex is developing and testing three continuous processes which make extensive use of cost-saving automated technologies, and hopes to concentrate all Naprosyn's production in one large efficient plant. Although this is an important strategy, Syntex will probably be unable to cut Naprosyn's price substantially. Because of the relatively small size of the market for pharmaceutical products, drug manufacturers cannot usually exploit economies of scale as extensively as in the commodity chemical business {CW, 1988c). Syntex is also patenting a number of different ways to fabricate Naprosyn. For instance, only one of the three processes under development uses Naprosyn's base compound, naproxen, as raw material. The company also claims it can manufacture Naprosyn using different feedstocks and processes. By patenting various Naprosyn processes, Syntex can make it harder for generics producers to manufacture the product after its patent expires (CW, 1988c). Moreover, Syntex is trying to improve Naprosyn. It has already introduced important enhancements, like a new controlled-release version and different formulations. Further, in 1988 it signed a deal with Procter & Gamble to market an OTC version of Naprosyn. Procter & Gamble has an extensive marketing network, which can be of great help for large-scale distribution, and can be an effective barrier against competitive products.
100
Case studies of US pharmaceutical companies
Finally, Syntex is offering to supply generics manufacturers with the basic chemical ingredients of Naprosyn so that they can formulate their own version of the drug. This is unlikely to prevent a fall of Naprosyn's sales after its patent expires. However, Syntex hopes to recover part of the forgone revenue by selling the bulk chemical products (CW, 1988c; BW, 1988e). These attempts to protect Naprosyn's position after 1993 are dictated more by the need to curb Naprosyn's decline than by aggressive competitive goals, like increasing Naprosyn's market share or, more generally, reinforcing Syntex's overall competitive position. Syntex is also trying to develop new drugs. In the late 1980s, it had a number of compounds waiting for FDA approval, e.g., Cardene, an anti-hypertensive, which is produced under license from Yamanouchi Pharmaceuticals; Ticlid, a stroke-prevention drug; Toradol, a non-narcotic analgesic for pain relief; Ketorac, eyedrops to relieve post-surgery eye inflammation; Gardrin, an anti-ulcer drug; and Cytovene, an orphan drug to combat a virus that can cause blindness in AIDS patients. Yet, these are not major new drugs which could boost Syntex's sales as much as Naprosyn did in the mid-1970s (CW, 1988c; BW, 1988e; BS&C, 1988). Syntex is implementing an aggressive strategy for in-house research. Its R&D to sales ratio has increased from 12.7 percent in 1983 to 17.1 percent in 1988, and it has been around 17.5 percent between 1989 and 1991 (Moody's International Manual, various years). The company appears to reinvest Naprosyn's profit in research, and it is trying to exploit its patented position to build up future competitiveness. Further, Syntex is setting the basis for sound in-house basic research. Among other things, it is building a new center for cancer research in Palo Alto, which will host advanced research equipment and qualified personnel for scientific work in this field (CW, 1988c). Syntex's story illustrates how a one-drug company is attempting to depart from its reliance on one major product. On the one hand, it is aggressively protecting its compound against competition from generic products after its patent expires. On the other hand, it is investing heavily in research to develop major new prescription drugs. Will this strategy succeed? This is still an open question. Syntex is clearly relying on research. But good in-house research capabilities do not necessarily translate into profitable outcomes. For one reason, even with sound research skills, pharmaceutical innovations still depend a great deal upon serendipity. More generally, the high costs, length, and uncertainty of drug research imply that Syntex faces high risks, and it is unclear whether it will be able to come up with an important new product. Syntex exhibits an important peculiarity with respect to the other drug companies examined in this chapter. It is really on the boundary between
American Home Products
101
being a "pure" research company and a "true" drug manufacturer with significant development and distribution assets. Syntex is the smallest among the "large" US drug companies (see table 4.1), and it is more renowned for its research (especially in contraceptives) than for its development and distribution assets. This suggests that a sensible strategy might be to reinforce its research capabilities, and aim at alliances for development and distribution. Clearly, this is not to say that Syntex should become a "research boutique," like many small to medium-sized biotechnology companies. But Syntex could make important choices about specialization. It could encourage even more strongly in-house research in the fields wherein it already has relatively greater strength. It could then take advantage of the emerging opportunities for a division of labor in drug innovation. In the end, Syntex might even benefit from acquisition of part of its equity capital by larger pharmaceutical groups, if it could maintain sufficient organizational and strategic autonomy, gain the necessary financial stability for research, and obtain access to extended complementary resources for development and commercialization.
American Home Products
American Home Products (AHP) is one of the largest pharmaceutical firms in the US market. Yet, it is not a major competitor in the research-based segment of this industry; nor does it have major in-house research capabilities. AHP has important presences in oral contraceptives and infant nutritionals. It has an extended line of OTC products, and it is active in the food and household business. Its performance in these markets is considerable. AHP's case shows that not all large pharmaceutical corporations base their strategy on solid in-house research bases, and some of them found their competitiveness on marketing capabilities and related assets in nonresearch-intensive businesses. In the late 1980s, AHP had some new drugs in the pipeline. Ultradol, a new anti-arthritic agent, was awaiting FDA approval. In the early 1990s, AHP expects to file an NDA for Tolrestat, a drug to treat the debilitating effects of diabetes. Tolrestat is already marketed in Italy and Ireland (BS&C, 1989e). In 1988, AHP acquired Robins, a pharmaceutical company for which chapter 11 of the Bankruptcy Act was filed because of lawsuits against its Dalkon Shield contraceptive, which had been responsible for infertility in women. Again, the acquisition of another large pharmaceutical company was a major step in meeting the rising competitive pressures of the drug market. However, the mergers between Bristol-Myers and Squibb, and
102
Case studies of US pharmaceutical companies
Smithkline and Beecham, sought to join significant research and marketing skills and assets, to form giant companies with considerable capabilities in the ethical drug business. Robins had a strong OTC product line, and it was not a research-based company. In spite of its financial difficulties arising from large legal compensation payments to women harmed by the Dalkon Shield, Robins still had good in-house capabilities in many of its product lines. AHP's acquisition was then not used to enter into the prescription drug market. It reinforced AHP's traditional businesses. More generally, AHP does not seem to be taking any significant steps to enter the highly competitive research-based ethical pharmaceutical market. Rorer
In the mid-1980s, Rorer had a modest research base. Its major operations were cosmetics and OTC medicines. It had an important OTC drug in the market, Maalox, the leading consumer antacid product in the USA. But it had practically no prescription drug line. In 1986, Rorer embarked on an aggressive strategy to enter into the market for R&D-based drugs. It acquired the ethical pharmaceutical business of Revlon, which raised its sales from 313.7 and 338.1 million dollars in 1984 and 1985 to 844.6,928.8, and 1041.6 million dollars in 1986, 1987, and 1988. Its R&D expenditures increased by almost four times, from 16.3 and 17.9 million dollars in 1984 and 1985 to 69.7, 81.8 and 102.8 million dollars in 1986, 1987, and 1988, i.e., a rise in the R&D to sales ratio from slightly more than 5 percent in 1983 and 1984 to above 8 percent in 1986 and 1987 and to almost 10 percent in 1988 (Moody's Industrial Manual, various years). Rorer also embarked on a program to focus on its most promising research lines. It concentrated research in five areas (cardiology, gastroenterology, hypersensitivity, bone metabolism, and hematology) (WSJ, 1987a). Rorer's strategy, although a courageous one, was also risky. Rorer needed to achieve a quick breakthrough. The cash flow from Maalox did provide the necessary funds to carry out Rorer's research restructuring. But Maalox, an OTC drug, did not generate a cashflowcomparable to that of a major prescription drug. An important success was necessary to break into the spiral of self-reinforcing advantages, namely the obtaining of a major boost in sales and profits to support further research on a long-term basis. Rorer was unable to produce a major new discovery in a relatively short time. In 1990, it was acquired by Rhone-Poulenc. The French company had undertaken a program of internal reorganization similar to Rorer. It had abandoned various business lines, including some of its traditional
Conclusions
103
concerns like textiles and fertilizers, to concentrate on research-based pharmaceutical products. Like Rorer, it had acquired various companies to expand its remaining operations (BW, 1990a). The merger with Rorer integrated various complementary assets, and it was explicitly aimed at reinforcing both firms in the research-intensive segment of the drug business. Rorer brings to the new company some of its internal research capabilities. Moreover, apart from its US sales force, which will help in selling Rhone-Poulenc's products in this market, it carries Maalox profits to sustain joint research. Rorer's research assets will pool with RhonePoulenc's research organization, built around its major research center, Institut Merieux, and the recently acquired Connaught Biosciences, a small biotechnology company specializing in vaccines (Bioscan, 1988; BW, 1990a). Conclusions: Lessons from the case studies We can draw three main conclusions from the case studies. First, although a public good, science is not a "free" good. Internal scientific capabilities are critical for taking advantage of the public good. Firms like Merck and Eli Lilly paid systematic attention to scientific research and they run research laboratories almost like academic departments. They have been more effective than their rivals in taking advantage of new scientific ideas, and in the 1980s they showed notable innovation and market performance. Squibb, Syntex, and Smithkline had good internal research operations, and they produced important innovations. However, their one-drug company status put considerable pressures on their performance. Bristol-Myers and Rorer had no long-standing research tradition. They reorganized their internal operations to gain research capabilities. However, although intensive and to some extent successful, these efforts were insufficient to sustain the competitive pressures of an industry in which solid intra-mural research expertise cannot be acquired even within a decade. American Home Products had no research tradition, nor did it attempt to establish one. It reinforced its non-R&D businesses, and it is not investing in the patented drug market. Second, the case studies showed that in this industry innovation and market performance are cumulative, and they can be a source of persistent heterogeneity across firms. The commercialization of important new drugs generates the cash flow that is necessary to sustain further, expensive longterm research.15 Merck and Eli Lilly used their proceeds from past breakthroughs to invest in new research. In the 1980s they developed a steady flow of new drugs. The importance of a sound financial basis from
104
Case studies of US pharmaceutical companies
previous successes is epitomized by Smithkline's story. Despite Tagamet's success, Smithkline did not reinvest its resources in scientific research. It forwent the opportunity of exploiting cumulative advantages, and faced poor innovation performance. As Rorer's case suggests, it is extremely difficult to build an extensive research basis from scratch, and Rorer was unable to break into the market of R&D-intensive drugs. Finally, the case studies hinted at some new trends in the market structure of this industry. Rising R&D costs imply that only giant corporations with formidable R&D, marketing, and financial capabilities will be able to afford extensive new drug developments and commercialization. In the 1980s we observed important consolidations: Bristol-Myers and Squibb merged; Smithkline merged with Beecham; American Home Products acquired Robins; Rorer merged with Rhone-Poulenc. Moreover, Sterling Drug, a major drug manufacturer, was acquired by Eastman Kodak in 1988; Merrell Dow and Marion Laboratories merged in 1989; in 1990 Hoffmann La Roche acquired 60 percent of Genentech, one of the few biotechnology companies that succeeded in becoming an integrated pharmaceutical manufacturer. Merck itself set important marketing and research deals with Johnson & Johnson and Du Pont, two other giants in the medical and chemical business.16 Does consolidation, and the formation of giant corporations, imply the demise of a division of innovative labor in the drug sector? The formation of giant companies is necessary to accumulate significant development and marketing capabilities to sustain increasing innovation costs. Very large firms can also accumulate considerable financial and organizational resources to perform, among other things, upstream research based on lumpy assets. Yet, this does not imply that they will forgo the opportunity of taking advantage of smaller specialist suppliers of research and ideas, especially if the latter have greater comparative ability in "invention." In fact, the very tendency towards establishing huge development and commercialization assets can be thought of as an attempt to strengthen downstream capabilities, and hence downstream specialization. This implies greater marginal benefits from relationships with agents endowed with complementary assets, and particularly with those that possess complementary abilities in upstream generation of research ideas. As Eli Lilly's story suggests, large drug companies are increasingly recognizing these opportunities, and they are taking advantage of external linkages in research. Moreover, a very important factor that will encourage large companies to exploit the benefits of a division of labor in innovation is that, as argued in the previous chapter, they can pick only promising research outcomes in the market, thereby moving the risks of early research stages onto their suppliers. If the new trend is really less bright for some, it will
Conclusions
105
probably be so for the medium-sized (or even large) drug companies that will be "caught in the middle." They may be unable to acquire (or build) solid assets for very large-scale development and marketing, or to specialize in research and produce ideas that can be matched with the complementary downstream resources of larger companies.
5
Scientific research and drug discovery: an econometric investigation
Introduction
It should now be clear, even to the reader with little knowledge of the pharmaceutical industry, that drug innovation can be broken down into two distinct stages: drug discovery, and development and commercialization. Discovery is the more creative step, but it is very far from sales. Although less ingenious, development commands considerable outlays of resources, and it takes place over a fairly long span of time. As they can be distinguished so clearly, we shall examine these two stages separately. This chapter focuses on the economics of drug discovery, whilst the next chapter looks at development and commercialization. As the previous chapters suggested, drug discovery depends on two factors. The greater the number of molecules tested, i.e. the greater the scale of applied laboratory research, the greater the expectation of finding a given number of compounds for clinical trials. Drug discovery also depends on the scientific capital of firms. The latter helps in discerning areas wherein researchers are more likely to find candidate drugs, thereby increasing the probability of discovery. This chapter presents a model of drug discovery in pharmaceutical companies. Discovery is assumed to stem from expenditures on two inputs. Each year firms invest in testing a number of molecules in their laboratories. This is the variable input, which can be thought of as applied research. They also invest in accumulating knowledge capital, which is the stock input, and it is proxied by the past scientific publications of firms. The number of discoveries is proxied by the number of company patent applications. The number of patents will be greater the greater the number of molecules tested. But a greater number of patents can also arise because firms search for new drugs among more prolific families of compounds. As discussed in previous chapters, this is precisely what is provided by scientific capital. Whether because it enables firms to produce scientific 106
A model of drug discovery
107
information internally, or to utilize external knowledge more effectively, scientific capital enhances the ability of firms to test compounds in areas with better yields. The model assumes that firms maximize the expected discounted stream of the benefits that they derive from their yearly patents, net of expenditures on testing molecules, and of the costs of investing in knowledge capital. Using a standard result in the literature, knowledge capital times its shadow price is proportional to market value. The model produces a patent equation and an equation for scientific papers, which are jointly estimated. In the patent equation, patents depend upon past scientific papers, and variables that proxy for the unit cost of testing molecules in the laboratory. The coefficient of the unit cost variable is related to the productivity of laboratory research, whereas the coefficients of past papers measure the productivity of knowledge. The second equation relates the annual flow of company scientific publications to market value and past knowledge capital. The model is estimated using data for the fourteen largest US pharmaceutical companies during 1968-91. (The firms are listed in the appendix to this chapter.) Other studies have estimated the relationships between drug company patents and their scientific papers. (See Koenig, 1983; Halperin and Chakrabarti, 1987; Narin, Noma, and Perry, 1987.) They all found positive correlations. This chapter, however, develops an economic model to interpret such correlations. Moreover, it relates the accumulation of knowledge capital to the expected future profitability of drug companies, as proxied by market value. Finally, it estimates the patent and paper equations in two sample periods, 1968-79 and 1980-91. We can thus assess whether there have been structural changes in the productivity of either laboratory research or knowledge capital, or both. The next two sections present the model and discuss the empirical results, respectively. The final section summarizes the chapter. An appendix describes the data. A model of drug discovery Search models typify the basic features of drug discovery.1 Laboratory research is like the sampling of balls from urns which contain black and red balls in different ratios. Urns are families of compounds. "Black" balls are inactive compounds; "red" balls are active compounds, which can be submitted for clinical trials. Applied chemists draw balls from the urns. Scientific knowledge provides information about the ratio of black to red balls in the different urns. It helps scientists in making more informed choices about the urns from which they can make their draws.
108
Scientific research and drug discovery
Define TV to be the number of draws of a given firm in a certain year, i.e., TV is the number of molecules tested in the laboratory. Define K to be the knowledge capital of a firm, and P the number of patent applications of a given firm in a certain year. Patent applications are a good proxy for discoveries in pre-clinical research. Because patents are an effective means of appropriating drug innovations, pharmaceutical companies patent their compounds as early as possible. Typically, they apply for patents concurrently with application for an Investigational New Drug Exemption (IND) to obtain permission for clinical trials. Patent applications are thus a better measure of pre-clinical discoveries than new drugs sold. Patent applications P increase with the number of molecules tested, N. They also vary with K, as greater K implies greater ability to search urns with better (expected) ratios of red to black balls. Using a multiplicative specification, the production function of patents can be written as P = A(t)NaKfie€
(1)
where all variables have subscripts / and t for firms and time, which have been suppressed for notational convenience; a and ft are the elasticities of TV and K. The function A{i) is a time trend that influences discovery. Grabowski, Vernon, and Thomas (1978) found that, other things being held constant, in the 1960s and the 1970s time had a negative effect on the number of discoveries. They attributed this result to depletion of research opportunities. The term e is a stochastic factor. Serendipity plays an important role in drug research (see, among others, Schwartzman, 1976). It is assumed that e = pe_{ + /x, where ^ is a stochastic term with finite mean and variance and uncorrelated over time, and />e[0,l] measures the persistence of serendipity over time, i.e., the extent to which "news" in previous years (e.g. unexpected discoveries of prolific families of compounds) influences present discoveries. It is also assumed that the accumulation of knowledge capital takes the following form K=XeKl~le
(2)
where Zis the number of scientific papers published by the zth firm in year t, and 6 e [0,1] is the percentage increase in knowledge capital produced by a unit percentage increase in the flow of papers. Papers are not an infallible measure of knowledge capital. The number of papers does not account for quality differences in publications. In addition, a large fraction of papers published by pharmaceutical industry scientists are in clinical research (about 45 percent of all papers published by this group; see Narin and Rozek, 1988). Clinical papers are typically statistical analyses of the effects of drugs in patients. They are likely to be more strongly correlated with the
A model of drug discovery
109
extent of clinical trials than with basic knowledge. (See, for instance, Spangenberg et aL, 1990.)2 Publications, however, are a common means by which scientific knowledge circulates. Thus, even though the number of papers may not provide a completely satisfactory measure of pre-clinical knowledge capital, it is correlated with the intensity of linkages of company scientists with the scientific community. As also discussed in chapter 3, pharmaceutical companies allow their scientists to publish to encourage solid ties with the scientific network, which provides advantages such as easier and more effective communication with academic scholars. Papers would then be a fairly appropriate measure of the type of scientific capabilities that one is trying to measure here. After all, what is really meant by internal scientific capital of firms in this discussion is the extent to which they are plugged into the outside scientific network. Relatedly, Halperin and Chakrabarti (1987) found a high correlation between the number of papers of pharmaceutical companies and the number of "elite" scientists that they employ. Elite scientists are people with a good reputation within the scientific community. The number of papers is thus correlated with an important measure of the breadth of linkages of company scientists with the scientific network.3 Pharmaceutical firms choose N, X, and K to maximize the present value of the expected discounted stream of utilities that they derive from their patents, net of variable costs in molecular testing, and the costs of accumulating knowledge capital, i.e. 00
max V0 = E0 £
[U(P)-CN-0(X,K)-X(K-XeKlSle)]r
subject to (1) and (2). In this expression: • subscripts / for firms and r for time have been suppressed for notational convenience. (From now on, I shall use subscripts - 1 and 1 to denote previous and next period.) • Vo is the market value of the firm at the beginning of the period, and Eo is the expectation operator given all information at the beginning of the period. • U() is the utility of each year's discoveries. One can think of £/(•) as the present value of profits that firms expect to obtain from future sales of innovations. I assume that U() is homogeneous of degree 1 in N and K, i.e. £/(•) is proportional to P~K • C is the cost of testing one molecule in the laboratory. • &(X,K) is the adjustment cost of knowledge capital, defined to be
whereof isyit averaged over time.
Table 5.2. Descriptive statistics: logs Variable
Mean
Std. dev.
Min.
Max.
Number of patents
4.34
0.54
2.40 (3.83)
5.76 91.9 (5.26)
53.8
Number of scientific papers
4.50
0.85
1.79 (3.14)
6.75 (5.67)
51.9
54.8
Sales/R&D
2.75
0.46
8.38
0.74
3.74 71.4 (3.71) 10.50 69.1 (8.89)
26.5
Market value (const. 1982 $m.)
1.70 (2.41) 6.43 (6.76)
BVAR (%) WVAR (%)
49.0
Notes: Fourteen firms, 1968-91. Number of observations = 303. Descriptive statistics computed using only observations in the estimated sample (see table 3.3). The sales/R&D ratio and market value are lagged one year (1967-90). In parentheses min. and max. for central year, 1979 (1978 for sales/R&D and market value) (fourteen observations). BVAR and WVAR are the ratios of the between-group and within-group variances to total variance, and they are computed as described in table 5.1.
114
Scientific research and drug discovery
scientific knowledge produced important changes in pharmaceutical research during the 1980s. One can thus test for changes in the productivities of laboratory research and knowledge capital. Relatedly, one can evaluate whether the lag structure of papers has changed. One reason for this is that, with greater scientific intensity in drug research, drug company papers may have become more "basic." This would mean longer lags before publication in the 1980s, and hence lower d.9 The very same fact that drug discovery has become more "scientific" would suggest lower p as well. With greater confidence in their ability to understand phenomena in "rational" terms, companies would be less inclined to pursue projects arising from unexpected and probably transient opportunities, and that are not grounded on solid knowledge. This was also discussed at some length in chapter 2 when suggesting that companies had less incentive to pursue different uses of molecules when information about other applications arose unexpectedly from clinical trials or from the market. In contrast, they may have greater stimuli today as they can rationally pursue different therapeutic applications of given compounds. Finally, by distinguishing between rj in the 1970s and the 1980s, one can test whether greater scientific knowledge and better experimentation technology have curbed exhaustion of innovation possibilities. The structure of the model also entails that differences in 6 in the two periods imply differences in y, 6, 0 l9 and 02> a s the latter are functions of 6. (See the previous section.) Particularly, y, i/tl9 and I/J2 increase with 0, whereas 8 decreases with 6. Hence, in allowing for the possibility that a, j3,0, p, and rj may be different, one must also allow for the possibility that all other parameters in (7) and (8) may be different. Estimation is accomplished by multiplying each term in (7) and (8) by a dummy variable which takes the value one for 1968-79 and zero otherwise, and by adding these expressions to similar expressions multiplied by a dummy equal to one for 1980-91 and zero otherwise. The results are given in tables 5.3 and 5.4. Table 5.3 reports the results for the constrained model. Table 5.4 presents results for the model with different parameters in the two sample periods. (In table 5.4, parameters have subscripts "70" and "80" to denote estimates for the 1970s and the 1980s, respectively) Table 5.4 also reports results from estimating the system (7)-(8) separately for the two subsamples 1968-79 and 1980-91.10 From table 5.3, a and /} are positive and statistically significant. However, their combined effect is not large. A 1 percent increase in both the number of molecules tested and knowledge capital implies only a 0.30 percent increase in the number of patents. As expected, companies need to test a large number of molecules, and they need to develop their knowledge capital a great deal in order to make just a few discoveries. The parameter rj is negative, which suggests depletion of research opportunities.
Empirical results
115
Table 5.3. Estimated parameters, equations (7)-(8) Parameters
Estimates
a
0.146 (0.036)
P
0.146 (0.064)
P
0.399 (0.085)
V
-0.020 (0.007)
d
0.861 (0.123)
y
0.181 (0.050)
8
0.542 (0.071)
0i
0.058 (0.013)
02
-0.001 (0.0005)
Notes: Fourteen firms, 1968-91. Number of observations = 303. Number of observations smaller than 336 ( = 14 x 24) because of missing values. Log likelihood function = -38.23. Heteroskedastic consistent standard errors in parentheses.
An interesting result in table 5.3 is the high value of 0. About 90 percent of the knowledge capital of firms in a given year depends on papers published in the same year. In other words, about 90 percent of the effect of the longrun elasticity j3 is explained by present papers, i.e. by scientific research performed two to three years earlier. Scientific research and drug discovery appear to be intimately connected. Many papers and patents probably deal with the same subject. While patents describe the molecules, papers elucidate the method with which they were discovered and other scientific details. The results in table 5.4 are even more intriguing. While a increases by
116
Scientific research and drug discovery
Table 5.4. Estimated parameters, equations (7)-(8) differences in parameters, 1968-79 and 1980-91
with structural
Estimates Parameters
(1)
(2)
(3)
0.060 (0.091)
0.022 (0.097)
0.238 (0.115)
0.262 (0.132)
P70
0.597 (0.219)
0.689 (0.277)
VlO
-0.031 (0.016)
-0.036 (0.030)
#70
0.902 (0.180)
0.895 (0.180)
770
0.230 (0.076)
0.231 (0.076)
^70
0.488 (0.108)
0.489 (0.108)
01;7O
0.066 (0.021)
0.066 (0.021)
02^0
-0.001 (0.001)
-0.001 (0.001) —
a
70
fto
80
0.192 (0.050)
#80
0.150 (0.094)
—
0.149 (0.094)
PSO
0.360 (0.118)
—
0.359 (0.118)
*?80
-0.021 (0.010)
—
-0.021 (0.010)
#80
0.687 (0.151)
—
0.684 (0.153)
780
0.087 (0.051)
—
0.093 (0.050)
^80
0.716 (0.086)
—
0.717 (0.087)
01;8O
0.044 (0.018)
—
0.045 (0.017)
a
0.191 (0.050)
Empirical results
117
Table 5.4. {cont.) Estimates
Parameters 02-80
(1) -0.001 (0.001)
(2) —
(3) -0.001 (0.001)
Notes: Heteroskedastic consistent standard errors in parentheses. Subscript 70 denotes parameter estimates for 1968-79. Subscript 80 denotes parameter estimates for 1980-91. (1) Fourteen firms, 1968-91. Number of observations = 303. Number of observations smaller than 336 (= 14 x 24) because of missing values. Log likelihood function = -32.58. (2) Fourteen firms, 1968-79. Number of observations = 154. Number of observations smaller than 168 (= 14 x 12) because of missing values. Log likelihood function = -34.80. (3) Fourteen firms, 1980-91. Number of observations = 149. Number of observations smaller than 168(=14xl2) because of missing values. Log likelihood function = 13.09. more than three times between the 1970s and the 1980s, ft, p, rj, and 0 decrease (in absolute value). The model exhibits internal consistency. The parameters y, tftl and i//2 ought to decline with 0, while 8 ought to increase with 0. Although no parametric restrictions were imposed, the estimated values of y, S, i/tl9 and \js2 vary consistently with the decline in 0. As reported in table 5.5, the statistical significance of inter-period differences in a, j8, p, 77, and 0 is only marginal. Also, from table 5.5, using the likelihood ratio test, one cannot confidently reject the null hypothesis that the models with and without structural changes are statistically equivalent. However, the discussion in the previous chapters suggests that the point estimates do capture some important changes in the evolution of drug industry research. The notable increase in a stems from two complementary effects. First, greater scientific knowledge implies that each draw from an urn is more likely to be a draw of a "red" ball: scientific knowledge provides industry researchers with information about more prolific urns, and they sample from those urns. Second, the technology of molecular testing has improved dramatically. Compounds can be tested via computer simulation and other sophisticated instruments, which has augmented the productivity of laboratory research. This implies that it is less costly to test compounds in the laboratory, and the unit cost of assaying one molecule (C in our model) has declined. In other words, with similar (real) expenditures in applied
118
Scientific research and drug discovery
Table 5.5. Hypothesis testing Testing of the estimated parameters, first column of table 5.4; s{-) - estimated standard error of parameter difference. Null hypothesis u
n
H o :a 7 O -a 8 o = O u.D
R -0
Test statistics (Student's t) a
7Q- G
G = H,M,L, where the c's are parameters to be estimated. They account for the impact of lagged average profits of the firms in group G on the sales of the ith firm, for G = H,M,L. One can then test for differences in the intensity of competition from the H, M, or L firms by looking at the estimated parameters H, M9 and L. Ultimately, the estimated sales equation becomes 8
sit = csil + akit_, + |97OD7O, X exp( - A 7 0 ) - [ A 7 Y 2 / ( T - 2)!]-rft_T + T=2
8
&0D80, X ex P (-A 80 )-[A 8 V 2 /(r-2)!]T,_ T + r=2 Z
H iHt- 1 + MZiMt- 1 + LZiLt- 1 + €it
(3)
where eight R&D lags have been used to estimate the ]8's and the A's.9 The second equation is the R&D equation. Assume that the (real) cost of innovation projects is an exponential function of Pit. R&D in real terms is the cost of undertaking such projects in a certain year. Then, Rit = rnOi-Qxp(mPit), with mOi and m > 0. Solve this expression for Pin substitute Pit in (1), and solve (1) for rit (recall rit = \ogR^. One obtains rit = const, + mtPit + yrit_ x + Sit
(4)
where dit has been normalized with respect to m. To derive the R&D equation to be estimated, an expression was specified for the optimal path of tPin i.e. the number of innovation projects initiated by the firm at time t. Pharmaceutical companies allocate funds to basic and applied research according to their financial capabilities, which depend upon present and past market performance. Assume that the optimal path of tPit depends upon log-sales in previous years, which capture the effect of past market performance, and the shock eit to sales and variable profits, which captures the effect of present factors on the decision to start innovation projects. The optimal path of tPit also depends upon firmspecific characteristics (e.g., different managerial attitudes towards research). Assuming linearity, one can write
Model of the innovation process
133
*it = Cri + yrit- 1 + €iSit- ! + &U-2 + f 3*,7- 3
where four sales lags are used to account for the effect of past market performance (fl9 £2, £3, and £4 are parameters); cri is a firm-specific (deterministic) component, and xe is a parameter that measures the impact of present sales shocks on the decision to start new innovation projects. The estimated values of x€ and the fs will test whether, as suggested by Grabowski (1968), Grabowski and Vernon (1981), and Jensen (1988), internal funds are correlated with R&D in the pharmaceutical industry. The proposed structure of the model suggests that internal funds influence the number of innovation projects tPit started at time t, i.e. they are an important determinant of expenditures in pre-clinical research. The third equation is the capital stock equation. It follows from the firstorder conditions of the optimization problem. The first-order condition with respect to capital stock at time / is E, p(d7Tit+xldK^ = ait. With Cobb-Douglas technology, variable profits are proportional to sales. If a is the elasticity of capital, one can write E,p-a-const-(S//+ JKit) = ait. Take logs of both sides. Substitute E^ l7+1 from the one-period lead of (3). Assume that Etcit+! depends only on cit, and not on eit_ l9 eit_2,.... Moreover, assume that Etf(Zij^ = ^'f{Zijt_^), i.e. the performance of competitors at time t is predicted by their performance at t— 1 (0 is a parameter). Solve for kit £l7), and obtain K = cut + / / V O " a)]D70, X e x p ( - A70)-[A7V2/(r- 2 ) ! ] T , _ T + , + T= 2
[/W(l - a ) ] D 8 0 , 1 exp( - A80)-[A8V2/(r - 2)!]-r,_ T+ , + r=2
[0/(\ - a)](0H-zH/ + ^M-zM, + L'ZLt) + y€-eit + ye-eit + 8, where ckit is the sum of firm and time components. The parameters y€ and yd measure the impact of the shocks e and 0 on capital stock; 8it is an error idiosyncratic to capital, which is interpreted to be a demand shock for non R&D-based products.10 The error components of (6) can be justified as follows.11 Sales shocks e affect investment in physical capital. News of sales raises the incentives of firms to expand their capital stock to satisfy the increased demand. ("Bad" news would work in the opposite way.) Technological shocks 0 may also influence the capital stock. Unexpected advances or failures during the innovation process may alter the propensity of firms to build new capital in anticipation of possible future sales expansions or contractions. Yet, given the length and risks of drug development, technological shocks are likely to be too distant from commercialization, and they can be very volatile.
(6)
134
The innovation cycle
Hence, companies could be very conservative in responding to signals during clinical trials. Drug firms would expand physical capital only when new products were almost ready to be marketed, presumably during NDA revision. This hypothesis can be tested. If drug companies change their capital stock only during NDA revision, the shock e will have a substantial impact on capital, whereas 6 will explain only a modest fraction of the variance of k. Empirical results Four equations were estimated. The first three equations were the sales equation (3), the R&D equation (5), and the capital stock equation (6). The fourth equation was a market value equation. Following Pakes (1985), under the assumption of stock market efficiency, deviations of market value Vit from its expected level, given all information at t—l, depend only on news at t, i.e. [(K z ,-E / _ 1 F // )/KJ = we€// + w ^ + wsS// + ^ 7 , where \fsit is a measurement error and the w's are parameters.12 If (Vit -Et_xVit) is small with respect to Vit, then [ ( ^ - E ^ F ^ / F J «log( Vit/Et_! Vit). Assume that the value of the firm is expected to grow at a rate qt common to all firms but different over time. Then, Et_xVit = Vit_ j Qxp(qt), and one can write l o g — ^ = qt + w€eit + wdSit + w88it + 0 , v
(7)
t-\
which is the value equation to be estimated. The sales equation (3), the R&D equation (5), the capital stock equation (6), and the value equation (7) were estimated using data on total sales (in real terms), real R&D expenditures, and a measure of the value of the capital stock in real terms, for the largest fourteen US pharmaceutical companies between 1968 and 1991. The market value of the firm is debt plus equity. Net income is equal to net sales plus other incomes minus costs of products sold, other expenditures, and taxes. Tables 6.1 and 6.2 report descriptive statistics. The appendix to this chapter describes the data. First differences were used in the sales equation (3), the R&D equation (5), and the capital stock equation (6), to eliminate firm-specific effects in csin cri, and ckit.n As their constant terms also have time-components, the dependent variables and the regressors in the sales and capital stock equations are differences from time means. Similarly, in the value equation, the dependent variable log(FJF, 7 _ 1 ) is measured in terms of differences from time means, and it is regressed against a constant.14 The system (3)-(5)-(6)-(7) was estimated by maximum likelihood, and the results are given in table 6.3.
Empirical results
135
Table 6.1. Descriptive statistics: levels Variable
Mean
Std. dev. Min.
Sales
2,672.3
1,534.8
R&D
194.5
136.1
2,289.4 Net value of capital stock
1,433.8
After-tax net income Market value
368.3
299.6
6,411.3 6,414.9
Max.
BVAR (%) WVAR (%)
195.6 9,174.3 67.3 (598.7) (5,358.2)
29.0
742.3 39.6 (246.0)
64.4
155.1 7,432.3 50.0 (618.0) (3,327.7)
58.2
20.9 (53.7)
9.9 (81.9)
2,078.1 49.6 (509.0)
62.7
622.1 56,954.1 54.8 (883.7) (7,013.1)
68.4
Notes: Fourteen firms, 1968-91. Number of observations = 277. Because of missing values, descriptive statistics are computed using only observations in the estimated sample (see table 6.3 below). Values in constant 1982 million dollars. In parentheses min. and max. for central year, 1979 (fourteen observations). BVAR is the ratio of the between-group variance to total variance, i.e. BVAR = Yjit(yit~~y*)2lYsit(yit~y*)2i w n e r e y* is yu averaged over firms, and y* is the overall average. WVAR is the ratio of the within-group variance to total variance, i.e. WVAR = ^ l7 (y /r —^f) 2 /^,-^—^*) 2 , whereof isyit averaged over time.
Table 6.2. Descriptive statistics: growth rates - log (yjyu-i Variable
Mean
Std. dev.
Min.
Max.
BVAR (%)
WVAR (%)
Sales R&D Net value of capital stock After-tax net income Market value
0.052 0.081 0.074
0.073 0.060 0.085
-0.187 -0.077 -0.264
0.550 0.330 0.530
75.4 71.9 74.1
87.7 93.5 87.8
0.067
0.357
-3.135
2.877
90.0
98.0
0.068
0.234
-0.608
0.777
55.1
98.0
Notes: Fourteen firms, 1968-91. Number of observations = 277. Descriptive statistics are computed using only the observations in the estimated sample (see table 6.3 below). Growth rates computed from variables in constant 1982 dollars. BVAR and WVAR are the ratios of the between-group and within-group variances to total variance, and they are computed as described in table 6.1.
136
The innovation cycle
Table 6.3. Estimated parameters, equations (3), (5), (6), and (7) Parameters
Estimates
Parameters
Estimates
a
0.064 (0.039) 0.111 (0.064) 0.664 (0.722) 0.459 (0.152) 2.816 (0.903) 1.399 (0.506) -0.755 (0.364)
0M
-0.489 (0.278) -0.079 (0.154) 0.638 (0.053) -0.040 (0.068) 0.041 (0.052) 0.119 (0.043) 0.095 (0.051)
£70 ^70
fto ^80
&> a n d £4 is positive. With constant growth rate of sales, past sales growth has a positive impact on R&D growth.15 As discussed in the previous section, this suggests that past market performance influences the decision of firms to start new innovation projects. The estimated parameters